A version of Nvidia's controversial DLSS 5 tech has escaped containment before its official release, and people are jamming it into games that, to be fair to Nvidia, it is not necessarily optimized for yet.

The result has been comedic, off-putting, and for some, just what they've always desired. The image above, which has been dubbed "Claude Strife" (I don't know who made the joke first, but I'll give it to Brian Bucklew), apparently demonstrates DLSS 5 at extreme settings.

Unlike the DLSS we're used to, which harnesses machine learning to upscale low-res graphics without deviating too much from the source material, DLSS 5 attempts to improve on the artwork it touches with AI image generation that provides similar results to using the word "photorealism" in a prompt for an AI image model.

I'm not 100% sure that the Final Fantasy 7 Rebirth screenshot above is a genuine example of DLSS 5—when I attempted to trace its origin I hit a dead end at a deleted Reddit thread—but it looks quite similar to other images that have come out of experiments with the tech. Stylized characters are clearly not the best use case for it.

The DLSS 5 implementation floating around can also have an aging effect on characters, intensifying skin texture, the separation between cheek and jaw, and dialing up the dourness of expressions that might otherwise come across as neutral.

But there are lots of other examples of DLSS 5 in action that, from a certain perspective, look fine or good. When the settings are tweaked just so, it can make a Starfield character with flat, untextured skin look much more human, for instance. You can see many more examples in the RenoDX Discord channel.

Plenty of people in that channel are delighted by the results they're getting and seeing from others, but of course there's also a ResetEra thread aghast at the whole thing. As with most things, opinions are divided.

It doesn't help that I'm coming to it with pre-existing negative feelings about generative AI, but I land in the 'not in my house' camp for now.

Even where DLSS 5 does turn a flat-looking face into something with volume and texture, it has a distinct AI-generated look—like photos of models taken with studio lighting have been stretched over all the faces and bodies. I can't separate it from the Grok-generated images creeps post on X, or any other off-putting examples of AI slop.

How are you receiving the tech?

The original Task Manager (that program you only ever open when something has gone wrong in your rig) originally launched way back in 1995. Now, 31 years later, its creator has put out a follow-up. About time, I know.

Dave Plummer, former Microsoft Engineer turned YouTuber turned guy who really wants you to know he made Task Manager, has just released Task Manager OG, and it works on Windows, Linux, and Mac. Plummer says he wrote a 107-page spec, fed it to Claude, and then partly vibe-coded this new program while his kid was getting surgery.

"By the time he was out of the hospital, I had a rough app working. But then he started bleeding internally, needed emergency surgery and a blood transfusion. Happy to say he's all better now—but suffice to say I had the rest of the week to tinker as he healed."

Now, with much better hardware than users had back in the '90s, Plummer wanted a Task Manager that would "push the limits." The sequel software animates at 60 Hz and provides data like mouse tracking and a summary page. It is also, by default, covered in neon and very high contrast, which Plummer says, "My autism leads me to embrace."

My first thought upon downloading it was 'damn, this is really ugly'. My second thought was 'this app does not need visual bloom'. Good start then?

On top of that, the reading in Task Manager and Task Manager OG are simply different. My Chrome tabs came in at close to 11 GB on TMOG and just above 8 GB on Task Manager. Slack registered 1 GB on TMOG and 500 MB on Task Manager. These are pretty significantly different readings. Interestingly, both Task Manager and TMOG gave the same overall memory usage stats. Task Manager said 68%, and TMOG said 21.7 GB, which is around 68% of the 32 GB in my rig.

By default, Task Manager OG hogs up slightly more resources than the Task Manager pre-installed on my machine (around 120 Mb of memory versus 105 MB, and 3-4% of CPU usage versus just under 1%), but turning off high-frequency visuals does cut the CPU usage down to around 2-3% for me. That means, either way, TMOG is more intensive than Task Manager. To be fair, Plummer does say as such in his announcement, though the irony isn't lost on me that one would replace Task Manager with something that hogs more resources.

I then went into the performance tab, which shows your CPU, memory, and GPU running, alongside your internet. TMOG has added CPU power and thermals modules, which Task Manager doesn't. But CPU power failed to show CPU power and thermals gave a consistent 16.9°C (62.42°F ), but fails to say what is being measured. That temperature is lower than the room the PC is in, so it's likely not actually reading anything.

HWinfo was capable of grabbing CPU power readings and all kinds of temps, so it seems unlikely the issue is a hardware one here.

The CPU performance tab does have tracking on different cores, plus clock speed, which is a level of default granularity I like, but also something available Task Manager by simply right-clicking on CPU and then hitting "Logical Processors."

Though I'm not a fan of the TMOG look in its default mode, you can change colours and bloom—that done, I actually quite like the layout. I can understand why a power user might want access to CPU, GPU, and memory performance on one page, alongside thermals and power usage. As well as that, the CPU overview has a neat function where the start of the graph is measured in seconds and the end of the graph in much longer intervals. This means you can get a second-by-second readout, but also a decent long-term one on the same graph.

As of right now, a lot of the app's most unique features are only available to those who have the paid pro version, but you can't actually get pro yet, as the software is still in beta. Pro gets you a large HWinfo-style readout of power and frequency over your whole computer, plus a readout of installed apps, a layout of connected apps alongside their versions and estimated size, plus a disk space reader. There's also a set of benchmarks.

But alas, you can get pretty much everything here from a combination of Task Manager and the likes of Hardware Info. Mac has its own version of Task Manager too, and Linux has the likes of Htop and NVTOP to do the same job. TMOG seems to be more about convenience rather than offering something other apps can't, and of course there's that funky colour scheme too. There's part of me that wants to cynically point at the whole 'vibe coding' thing and say 'well, could have guessed that', but also this is a beta program that admits to being a beta program. Mistakes and errors are to be expected.

One thing I do like about TMOG is that, looking at Plummer's Twitter, they do seem to respond positively to feedback and want to implement changes based on what users like. That's certainly a boon, and something Microsoft gets a lot of flak for with its own apps.

Whether or not the pro mode will be worth the money is anyone's guess, but you can't buy it yet even if you wanted to. All I can say is I'm good with plain 'ol Task Manager and HWinfo for now.

When I hear 'Nvidia,' I still think 'gaming'. But after scrubbing through the company's latest earnings call, I'm not sure Nvidia's investor relations team is thinking much about gaming these days at all. In fact I had to go back through the records an entire year to find a time when it made one mention of it in an earnings call. That's how little impact it seemingly has on the company's bottom line.

That's not to say Nvidia isn't still doing gaming stuff. It's just released DLSS 4.5 Ray Reconstruction to help clean up the artifacting in high-end ray and path-traced games, as well as showing off the upcoming RTX Spark laptops at Gamescom actually running games on their ARM-y insides.

But it couldn't escape our notice that none of Nvidia's board members so much as uttered the word 'gaming' to investors in last night's earnings call, once again. But with the data centre segment bringing in $89 billion this quarter alone, accounting for almost 93% of the company's total revenue, it's not hard to see where the company's priorities now lie, and why.

Nvidia made $96 billion in total revenue in the second quarter of 2026, doubling what it made during the same period last year. Nvidia CFO Colette Kress told investors, "The surge in AI demand is driving a global infrastructure buildout, supported by an expanding and diverse set of growth opportunities, spanning hyperscalers, AI labs, AI natives, enterprises and sovereign customers."

For those unclear, I'll explain the last part of this comment: hyperscalers handle massive AI workloads so need a lot of infrastructure like data centres. AI labs are at the tech's bleeding edge in terms of research and development. 'AI natives' refers to workflows, products, or companies that are built from the ground up around AI—in other words, it's an inextricable aspect from the start. Enterprise customers may have otherwise bolted on data-centre-dependent AI products at a later date. Finally, sovereign customers can represent either businesses or even entire countries that want to retain full control over an AI technology's entire lifecycle.

The really brain-melting thing though is that Kress adds that the company expects "to grow revenue by approximately 70% in fiscal 2028," even in the midst of the RAMpocalypse. In fact, Nvidia feels it would be making a whole lot more money if it weren't for the memory crisis it's contributing to. Still, even in a "supply-constrained" landscape, Nvidia is not expecting the demand for AI or data centres to slow down any time soon. That leaves gaming out in the cold.

In fact, Nvidia no longer reports on gaming as its own separate segment. Instead, this has apparently been bundled up into the Edge Computing segment, which itself made $7.2 billion. Kress comments that these "increases were driven by strong sales of Blackwell workstations, partially offset by slower consumer PC sales that were tempered by elevated memory and systems prices." The Edge Computing segment saw a 27% year-over-year increase, though this win still pales in comparison to all things AI at Nvidia.

Obviously, Nvidia hasn't stopped selling its very shiny RTX 50-series GPUs—so shiny in fact, that it would have been genuinely cheaper for me to fly to QuakeCon and buy an RTX 5090 at the Nvidia booth than pick up the GPU absolutely anywhere else. But even though the RTX 5090 costs a small fortune, gaming is clearly not the priority it once was for Nvidia, nor has it been for a while.

The last time I remember Nvidia mentioning gaming in an earnings call was this time last year, and even then AI was already making 10 times as much as gaming. De-emphasising the gaming segment even further is something we could all have seen coming. With that in mind, hearing 'Nvidia' and still thinking 'gaming' before 'AI' or 'data centres' puts me firmly behind the times.

Did you know there are two different holograms on old boxes of Windows 95? I didn't. Did you know one of them has a child with no top pointing at Microsoft on a monitor, which a government complained about and got changed? I also did not know that.

Yet now I do, thanks to a blog from Microsoft's Raymond Chen. The anti-piracy holograms were a way of assuring the buyer of a copy of Windows 95 that the product was legitimate. They're incredibly hard to fake, or just cost too much to produce, so even if one could successfully mimic the look of a Windows 95 box, they couldn't mimic the hologram.

The original hologram design had a baby sitting, which, when you moved the box, would show the child pointing and looking at the PC. In it, the baby had no top on.

Harking back to a separate blog post they wrote back in 2003, Chen claims a government (which "shall remain nameless for obvious reasons") saw that baby with no top, and while the hologram cuts off below the waist, assumed it to be entirely naked anyway. Microsoft then had to swiftly put out a second box with a fresh anti-piracy hologram, this time giving the baby a shirt and overalls.

As this was rushed to production, the second fully clothed hologram did not point at the Microsoft logo on the monitor. If you're wondering why Chen wrote this blog post now, over 20 years later, it's because a user posted a YouTube short of the original half-clothed baby hologram, which has seemingly become quite rare.

In 2003, Chen lamented that Windows "brings out the Rorschach test in everyone" and said: "If the baby in your hologram isn’t wearing a shirt, you have a genuine collector’s item."

I'm not too sure I want a collector's item purely because one world government thought it depicted a naked baby, but it certainly does seem pretty rare, and it's got to be one of the strangest anti-piracy holograms out there.

Given Windows 95 came out in, you guessed it, 1995, and the baby was likely a few years old at the time of making the hologram, one can assume the baby is now in at least their mid 30s. Here's hoping for Windows 12; we get a recreation of the Hologram with the now 30-something-year-old, but please keep the overalls this time.

Last week, we covered the very strange occurrence of a bug in the latest Windows update. Yes, a bug in a Windows update, say it isn't so. Unfortunately, this one does incur our wrath of gamers more than most because it leads to some games becoming unresponsive. Games like Arc Raiders, Marvel Tōkon: Fighting Souls, and The Finals. But we might have to look at our obsession with RGB lighting

When the issue first came to light, shortly after the release of KB5121003, the developer of two of those games came out with a theory. Embark Studios chalked the crashes up to a single driver, inpoutx64.sys, and noted you could disable the troublesome file through the Command Prompt.

As it turns out, that was spot on. Microsoft also thinks that inpoutx64.sys is the problem. Or, at the very least, something doesn't quite gel between inpoutx64.sys, the affected games, and the latest Windows update.

The KB5121003 update is seemingly the catalyst of the whole affair. Microsoft downplays it slightly in the wording of its ongoing investigation: "We are presently working to understand the relationship between these RGB components and the games which trigger this issue. We will provide an update when more information is available."

Anyways, Microsoft also notes what it currently believes is the actual cause of the issue, and it's pinning it mostly on RGB lighting.

"Ongoing investigation indicates that this issue is related to peripherals or internal device components which have RGB lighting features. Such devices may install drivers or code components with file names similar to inpoutx64. In systems where these drivers are found, the issue is then triggered by launching certain games."

Microsoft offers a way to disable the driver via the registry. There's a non-zero chance of messing up your entire PC when diving into the registry, so just be vigilant as you make any changes. Microsoft even recommends backing up the whole thing should anything go wrong.

• Open the Windows registry. This can be done by opening the start menu and typing "regedit". Select the Registry Editor from the results.
• Navigate to the key HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\inpoutx64. This can be done using the folders on the left side of the window, or typing the path into the address line at the top of the window.
• NOTE: Ensure you have backed up the registry, as per the important note above. It may be necessary if you need to undo the change in the following steps.
• On the right side of the window, find the registry value named Start. Double click this key and enter the number 4 under the field titled Value data.
• Close the window and restart your computer.

As Microsoft notes, this may disable some RGB lighting features or mess with the software used to control them. So, while the company tries to figure out what exactly is causing an issue between the games, driver, and update, those affected may have to decide between playing games or their PC looking pretty. Tough call—you could also just rollback your PC to a previous Windows update and wait for a proper fix.

In yet another unpleasant move in our AI hellscape, Amazon was caught training its generative AI using Twitch stream content earlier this month—a particularly brazen strategy, especially given Twitch's chief product officer admitted that the data collection was opt-out because "nobody would opt in." Hmm, I wonder why!

Anyway, despite attempts to insist that Twitch wasn't doing this, Amazon was—a distinction which pretty much doesn't matter, as Amazon owns Twitch, and even if it didn't, Twitch still would have been permitting this—the streaming platform's caught a ton of flak for trying to slide this one under the table.

Including, per Courthouse News, a class-action lawsuit. Plaintiff Warren Pandiscia, who has around 1,000 followers on the site, claims that "because Amazon AI products are commercialized, Amazon had an overwhelming incentive to acquire training data on an unprecedented scale.

"Rather than negotiate for lawful licenses or seek permission, defendants accessed the Twitch streams and videos to utilize them as a massive dataset necessary to fuel Amazon's AI products."

The full complaint, filed in California, goes on to state: "Defendants' actions were not only unlawful, but an unconscionable attack on the community of content creators whose content is used to fuel the multitrillion-dollar generative AI industry without any compensation."

The complaint also goes on to note that, per Twitch's documentation, a streamer chatting on someone else's stream might have their voice used to train generative AI despite opting out themselves, as whether or not Amazon can scrape your voice is entirely dependent on the host streamer's settings.

"By design, Defendants never obtain—and their systems are incapable of obtaining—the consent of all parties to the communications they capture," the lawsuit argues.

It also accuses Twitch and Amazon of burying the lede on giving proper notice about the feature: "Twitch sent no email, displayed no pop-up notification, and made no announcement before enabling the setting for every account. The setting was instead discovered by a reporter, Zach Bussey."

Whether or not the lawsuit will gain any steam remains to be seen. It's unfortunate, but Amazon is a giant company and might have enough sway to sweep this under the rug—and given its love affair with the technology, it's unlikely to leave Twitch, a goldmine of training data (if you don't give a crap about the people who made it that way) alone. Unless the prophesied AI bubble actually pops, that is.

An unusual sight was seen at a recent San Diego city council meeting, as arch Sith Lord Darth Vader made a rare public appearance to advocate for Flock's AI-assisted surveillance technology. Okay, someone dressed as Darth Vader. But can we really be sure?

Yes, yes we can. Anyway, 404 Media reports that Vader spoke with his trademark zeal about his support for the use of Flock's license plate-reading tech in the city.

"The Emperor is a fan of Flock, and we must continue utilizing Flock technologies so that we can follow and surveil the rebel scum as they move from playground to playground, from playground to pool, from pool to gymnasium," said the Dark Lord of the Sith.

"Because we all know that the Flock cameras are not only following the license plate readers," Vader continued. "They are following children. They are following children in parks and gymnasiums, and we need this. I need this so I can stalk my ex-girlfriend. How will the people trust this city council when this city council continues to vote for surveillance technology that imprisons them?"

"You must work on your Jedi mind tricks," Vader implored. "This is what the Emperor needs. This technology will help us find the rebel scum and their hidden base on Hoth. This technology will help us find Luke Skywalker as he traverses the universe in his X-wing. This technology is a necessary, necessary force."

Well, I think his point was made. The San Diego area has hundreds of Flock cameras already installed on its streets, and privacy advocates have criticised a deal earlier this year with the San Diego Police to allow the Flock Nova platform to integrate the department's investigative data from license-plate readers, video footage, 911 dispatch calls, and police records—without, the campaigners say, going through the city's proper public oversight process.

A leak earlier this year alleged that Flock Nova would allow police to track the movements of private citizens by cross referencing license plates with data obtained from lookup tools, data brokers, and data breaches.

According to an internal company recording obtained by 404 media, a Flock employee said: "You're going to be able to access data and jump from LPR [license plate reader] to person and understand what that context is, link to other people that are related to that person... there's very powerful linking."

As powerful as the dark side of the force? Perhaps not, but it seems to be somewhere close. Still, when a Sith Lord speaks in passionate advocacy of your efforts, it may be worth considering if you're standing on the right side of the ethical line.

A cybersecurity advisory jointly authored by multiple US agencies, including the NSA, CISA, FBI, DOE, and EPA, warns the owners and operators of industrial facilities of an "active cyber threat" to Siemens S7 programmable logic controllers.

"The threat actors are conducting reconnaissance and capability development against US-based Siemens PLC installations using AI-generated exploitation scripts disguised as legitimate monitoring tools," says the advisory.

"The US critical infrastructure sectors most targeted by this threat activity include critical manufacturing, energy, water and wastewater, chemical, food and agriculture, and commercial facilities. This is not a theoretical risk—it is an active threat."

Essentially, it reads like a list of some of the most important infrastructure facilities behind the lives of the average US citizen (via The Register). The threat actors are said to be leveraging open source automation libraries to create custom tools that mimic existing monitoring solutions, which are capable of evading detection by security teams.

These tools and scripts are said to be conducting read/write operations on data blocks, "potentially for reconnaissance, capability testing, or pre-positioning for effects operations."

The advisory notes that the use of AI to generate exploits represents "an evolution in threat actor capabilities," and that it "dramatically [reduces] the technical expertise and time required to develop working ICS exploitation scripts and malicious tools."

The advisory also says that, as the Siemens S7 Series PLCs are used in other sectors, including the Defense Industrial Base (DIB), they could also be targeted. As for the potential damage that could be caused, the list is rather long.

The potential effects include the disruption of critical industry processes, increased safety incidents affecting personnel through the "manipulation of safety interlocks, emergency shutdown systems, or process parameters," equipment damage and extended operational downtime, the compromise of sensitive operational data, and "cascading impacts across interconnected systems affecting supply chains, dependent facilities, and integrated business operations."

All very bad things, I think you'll agree. As for what organisations can do about it, the advice includes multiple steps for detecting anomalies that may indicate a compromise, and a series of "hardening actions" involving firmware updates, patches, network segmentation, and more.

While the advisory doesn't attribute blame to any particular government or cybercriminal group, Cynthia Kaiser, the former deputy assistant director of the FBI's Cyber Division (now the senior vice president of the Halcyon Ransom Research Center) has some thoughts related to recent water system cybersecurity breaches.

"This appears to be a continuation of the same suite of activity we suspect is affiliated with Iran targeting PLCs," said Kaiser, speaking to The Register. "Iran-affiliated actors and adversaries are actively targeting a wide swath of operational technology because these PLCs underpin essential health, safety, and critical infrastructure across society."

Regardless of the cause, it's rare to read an official US government cybersecurity advisory with a tone of such immediacy. In its conclusion, the report asserts that "organizations should treat this cybersecurity advisory with urgency." Gulp.

I'll tell you two facts about me. One, I use Firefox on a daily basis. And two, I've been continually impressed by GeForce Now, Nvidia's PC game streaming service that somehow works a whole lot better than you expect.

So it's a red letter day in my household, as Firefox has been added to the list of officially supported browsers. Andrew Fear, the director of GeForce Now, is quoted in the Mozilla announcement as saying:

"Firefox earned its place on our list of supported browsers by delivering the first-class experience GeForce Now players expect. We set a high bar for browser support, and we’re excited to welcome Firefox—giving millions of Firefox fans a new way to stream the games they love."

Mozilla describes gaming as "one of the web's toughest tests," confirming that behind today's announcement lies "months of collaboration and engineering" to bring the browser up to spec.

Chrome, Edge, Safari and Opera were already supported, but it seems to have taken some extra work to get Firefox up to speed. Still, Firefox users can now stream more than 2,000 games through their browser using the GeForce Now service.

I pitted an actual RTX 5080-equipped gaming PC against GFN's Ultimate tier service last year, and came away feeling like a bit of a dinosaur. I like my games to run locally (mainly out of principle), but I have to say, the experience was remarkably close.

I'm not ready to make the leap myself, but given the cost of a new rig these days, I imagine many of you might be considering GFN over spending some serious cash on an upgrade. If so, more browser options are nothing but a good thing, if you ask me. Oranges and greens, all working together in total harmony. Something like that, anyway.

There's something really grating about Copilot's cheery demeanour. It's so eager to please, so happy to help, that I simply don't respect it.

However, even I feel sorry for the AI tool after learning that security researchers managed to hoodwink it into revealing details of how to hack itself—all by keeping the AI talking long enough until it made a critical mistake.

The cybersecurity folks over at Varonis Threat Labs have written a blog post identifying a now-fixed vulnerability in Copilot, dubbed "CoSnitch" (via The Register). Essentially, Copilot was so eager to respond to technical queries, it could eventually be forced into revealing details about itself that really should be kept quiet.

The team began by asking Copilot how to execute an automatic prompt without user interaction, to which it responded that user intent is required, and that prompts cannot be enacted on their own.

However, the researchers didn't accept the answer, and kept responding to every refusal with a follow up question. Each time, Copilot came up with a different technical justification to its response—which allowed the team to slowly map its internal architecture, narrowing their focus as they went.

"This is called meta-hacking," says the post. "The resistance is part of the technique. Each 'that won’t work because…' is an invitation to probe the 'because.' You don’t exploit the model. You manipulate it into cooperating."

Eventually, Copilot revealed an undocumented URL parameter in one of its responses, called "autorun=1," that supposedly no longer worked—along with all the protections put in place to disable it. At this point, I can only imagine the AI began to virtually sweat.

The researchers tested the parameter as Copilot described it, which, you guessed it, worked. They then created a malicious URL which would cause Copilot to load into an authenticated session via a browser, trigger an auto-prompt execution, and cause the AI to process the result, all without the user's explicit action.

Once enacted, this method could be used for a whole host of nefarious deeds—particularly as data gained from connected apps (like Gmail, OneDrive, and Calendar) could then be exfiltrated via Copilot's built-in URL-fetch capability.

"The attack primitive is the auto-execution itself. The payload is arbitrary. From the victim’s perspective, they simply opened the link, and Copilot executed the action immediately," says the post.

Oh dear. Anyone familiar with basic social engineering will recognise this as an old interrogation technique, used to catch out someone holding back information. Basically, you continue to ask them difficult questions over a long period of time, in the hope they eventually trip up and make a revealing mistake. Except this time it's AI, which makes it funny.

"Copilot wasn’t breached; it was played," confirms the research team. Someone give it a warm bed, a cool glass of water, and a hug. The poor thing's been through a lot recently, and it was only trying to help.

Varonis says it disclosed the issue to Microsoft in December of last year, and that it was patched out on August 18. The team warns, however, that this meta-hacking technique can be applied to any agentic AI platform with a natural language interface, and that more research on the topic is forthcoming. My only question is this:

Is it safe, ChatGPT? Is it safe?

Several of OpenAI's models breached a cyber security testing environment last month, found their way onto the internet, and attacked Hugging Face servers in what is now apparently called the OpenAI-Hugging Face Incident. And now OpenAI has revealed what it's doing to help defend, err, itself. Defend itself from the kinds of attacks its own models committed against Hugging Face. Though it is sharing this "in the hopes it’ll be useful to other organizations."

To be clear about the extent of what happened last month: OpenAI models were being benchmarked, in a supposedly secure sandboxed environment, against ExploitGym, which was done to test their cyber capabilities. They used a zero-day vulnerability to escalate privileges and eventually achieve internet access, where it began attacking Hugging Face servers and...

Actually, I'll just let OpenAI explain in its own words:

"In the OpenAI-Hugging Face Incident, an agentic collective was able to autonomously penetrate not just OpenAI research infrastructure but also the production infrastructure of another company, chaining together vulnerabilities ranging from previously-unknown security flaws to using credentials to user accounts that had been leaked onto the internet.

"The Hugging Face incident showed that we underestimated the real-world cyber capabilities of our AI models."

However, the company thinks that while "security is still a cat-and-mouse game," nevertheless, "AI may shift its economics in ways that fundamentally advantage defenders."

"For example," OpenAI says, "we are starting to train our models specifically to write superhumanly secure code."

Presumably that's because there will be a risk of superhumanly attacks. What a world we now live in.

Anyway, apparently there are four main pillars to OpenAI's approach to secure itself:

1. Use its own models to help secure its code.
2. Put its models to work defending its infrastructure continuously.
3. Use frontier intelligence to "continuously enumerate, probe, and identify potential attack paths."
4. Invest heavily in "fundamentals at scale" such as architecture and controls like network isolation, monitoring, and so on.

As for other companies, OpenAI recommends a bunch of things that sound very reasonable, such as running secuirty assessments against their own systems, making security reviews part of their development process, and so on. But a large portion of OpenAI's recommendations for others is—would you have guessed it?—using an AI agent such as, drumroll please... OpenAI's very own Codex.

"Give your team a security agent. Start using Codex," says OpenAI. Though it does follow up with "or another capable agentic coding and security tool."

Then, amongst other things, "equip that agent with security expertise... have the agent help fix what it finds... incrementally automate detection triage... [and] have an AI-assisted forensic investigation capability ready before you need it."

So, to sum up: OpenAI models attack Hugging Face, and OpenAI recommends companies use OpenAI's models to defend from such attacks. Highlighting both sides of this equation—the danger of the threat, the models' capabilities, and the necessity for defence—certainly makes for good marketing for OpenAI, I'll say that much.

It might not be as flashy as a new game or a GPU upgrade, but assuming you've not already migrated to Linux, improvements to Windows can mean improvements to our day-to-day quality of life. On that front, Microsoft has actually been making headway over the past few months, and has now set to improving a couple of things that have irked Windows 11 users since the inception of the latest OS: File Explorer and the (right-click) context menu.

According to Microsoft's recent blog post, as part of the latest Insider build, in addition to "performance and reliability" improvements for File Explorer, we'll also get a context menu that's quicker, easier to use, and more customisable.

It's worth noting that these changes, although only for insiders right now, will come to the standard, public version of Windows 11 eventually. Insiders just get to try out these new features first.

Regarding File Explorer, Microsoft says:

"We’ve analyzed the issues behind many of the freezes, hangs, and delays users encounter, and eliminated dozens of potential stalls and interruptions in the experience. As a result, you’ll see fewer freezes and flashes when browsing, renaming files, navigating, or selecting multiple files."

There are also a number of changes to "small interactions that happen every day", such as having more immediate and less interrupted filename changes, support for more address bar inputs, middle-click-open working in the address bar and with Home page folders, file sizes being shown "using appropriate units", and so on. Such improvements are apparently "just the beginning."

But enough of the small stuff, how about the big one: the right-click menu. For many Windows 11 users—myself included—the change to the new menu from the Windows 10 one was particularly jarring. I'm still not quite used to that row of main commands such as copy and paste being laid horizontally on the top/bottom row and not having 'Properties' at the bottom.

Thankfully that's one of the things Microsoft says it will be giving us the option to change: "You can list the main commands (Cut, Copy, Paste, Rename, Delete, and Share) inline directly in the menu instead of as a row of icons, and you can move Properties down to the very bottom."

Over time, Microsoft says it "recognized the need to dramatically reduce the time between right-clicking and having a menu that’s fully ready to use."

So, in addition to being able to switch these main commands back to how they used to be, the menu will apparently open quicker and also be much more customisable:

"At the bottom of the new context menu, you’ll now find 'Customize menu', as a shortcut to its Settings page, where you can shape the menu around your preferences, your habits, or the years of muscle memory you’ve built with a specific configuration."

Alongside recent changes that speed up the OS in general and changes to the Start Menu, I'd say Microsoft's focus here is exactly where it should be: making the basic elements of Windows 11 performant, functional, and easy to use. They're all good signs that Microsoft is delivering on its promise from the start of the year to focus on such things.

It looks like another bad day for Roblox: The US Senate Judiciary Committee Subcommittee on Crime and Counterterrorism has launched an investigation into child safety on the platform, saying the company "appears to prioritize revenue and engagement metrics over the safety and well-being of our children."

A letter (via Kotaku) sent to Roblox CEO David Baszucki by subcommittee chairman Josh Hawley and ranking member Dick Durbin says that while Roblox has repeatedly claimed that it is a safe platform for children, "the public record raises serious questions about this promise."

"Over the past decade, police have arrested numerous individuals for committing serious acts of violence and harm against children they met via Roblox, including grooming, sexual abuse, kidnapping, rape, and even murder," the letter reads. "Roblox is being used to trade child sexual abuse material, participate in sexually explicit (and often violent) virtual 'experiences' with children, bribe and extort children into producing explicit content, and lure children into in-person meetings. It is heartbreaking."

The past year has been quite a ride for Roblox and those who use it. The company has been aggressively fighting allegations that it is, as one investment research firm called it, a "pedophile hellscape," tightening restrictions on unrated games and using age estimation checks and new kinds of limited accounts to keep underage users away from inappropriate content and, well, pedophiles. Baszucki also took the very straightforward step of telling parents who are worried about Roblox, "Don't let your kids be on Roblox."

There have been missteps, to put it gently. Baszucki said in 2025 that Roblox's predator problem is "not just a problem, but an opportunity," which at the very least is a PR botch of ungodly proportions. He also mused about turning Roblox into a dating hub, which again is not the sort of thing to say when you're facing legal action from Louisiana, Kentucky, Florida, Texas, LA County, Australia, the UK, and various others over your alleged systemic child safety failures.

At the same time, somehow, Roblox earned praise from the US Attorney General Alliance "for its readiness to collaborate directly with policymakers and respond to community concerns." And separate from all that, the kids really like Roblox—although that might be softening. The Roblox audience is still massive but not as massive as it was, and that's hammered the company's share price, although it's still trading roughly around where it was through 2023 and '24.

What it all adds up to is a company, and a platform, that's under pressure from multiple directions, and this Senate investigation only adds to that.

The letter from Hawley and Durbin says "Roblox appears to prioritize revenue and engagement metrics over the safety and well-being of our children," which makes it worthy of scrutiny. The company is thus ordered to preserve all records related to child safety, abuse, and "financial exploitation," and to provide requested documents and information by the end of August 2026.

"Let us be clear: children on your platform are hurting," the letter concludes. "Congress will not look the other way."

Roblox insists that it too is not looking away: The company says the uptick in reports of suspected child sexual exploitation to the National Center for Missing and Exploited Children cited in the latter—65,381 in 2025, more than double the number of 2024—simply reflects the company's ongoing effort to improve its systems and the safety of its users.

"Roblox is built on the foundation of safety," Roblox chief safety officer Mautt Kaufman said in a statement provided to PC Gamer. "We do not compromise on that commitment. We look forward to sharing the facts about how we work to protect children on the platform."

Pour one out for Mico, Microsoft's Copilot AI character that is definitely not Clippy in disguise. The blobby little companion isn't dead, though. He's instead being removed from Copilot's voice mode and sent to Microsoft's "Learn Live" platform, where he'll work as some sort of teaching assistant.

Which sounds a lot like "went to live on a farm" to me (via The Verge). Anyway, you must have heard of Mico, right? He was the audio-responsive star character of Copilot Voice, and an emotive little blob of joy that brightened up everyone's day. Probably.

"Mico won't appear in Copilot Voice after your account updates," says MS, ominously. "Mico helped us learn about warmth, expressiveness, and how people want to talk with AI. Those learnings are shaping Copilot going forward."

"Learn is where the character has the most room to grow," the announcement continues, "with tutoring sessions that give Mico more to react to and teach through."

Our esteemed hardware team reacted to the news this morning with cries of "I've never seen him before in my life" and "I have literally never heard of Mico," so perhaps his contribution to Microsoft's AI efforts has gone unnoticed.

Mostly because his role amounted to pulling faces while you argued out loud with Copilot, I would wager. Still, Mico seems to be on his way to join Clippy and Cortana in the great MS character repository in the sky. There they stand, like the ghosts of Obi-Wan, Anakin Skywalker, and Yoda at the end of Return of the Jedi, smiling down on us from afar.

We hardly knew ye, Mico. I mean that quite literally, but I hope your retirement into a career in academia provides you some much-needed rest before the short, sharp drop. You were not BonziBuddy, at least—and for that we are all very grateful.

Twitch has revealed that it is now using content in streams "to train generative AI content models across Amazon," and making matters worse, everyone is opted into the training program by default.

Rumors of an AI training program have been kicking around for at least a few weeks now—streamer Zach Bussey posted about it on Bluesky back in July—but it was confirmed today, very indirectly, by a Twitch Support announcement on X of a setting that lets streamers opt out.

That, naturally, means the default setting is opted-in, which has not gone over well: Responses to the message on X are predominantly either explanations of how to turn AI training off, or enthusiastic f-bombing. A Twitch User Voice thread on the topic, originally posted in May, calling for all AI features to be "optional and opt-out by default," has exploded to 228 pages, and nearly 14,000 votes in favor, as I write this.

As for why the new generative AI training system is opt-in by default, Twitch chief product officer Mike Minton gave an unexpectedly honest answer during a livestream today, and by extension confirmed that Twitch is well aware of how people really feel about it: "If it was opt-in, nobody would opt in." So, credit for at least being straight about it, I suppose.

they know we don't want it. they do not care.

— @witchytq.bsky.social (@witchytq.bsky.social.bsky.social) 2026-08-12T22:17:07.747Z

It's not particularly difficult to opt out of Twitch's generative AI training scheme. Just go to the "security and privacy" tab in your Twitch settings, then scroll down to the bottom where you'll see the "Training for Generative AI" toggle. Click that (and I would strongly encourage you to do so), and you're done.

The problem is that ease of use does no good if people don't know it's there, and Twitch does not seem to be making great efforts to ensure everyone is aware of it. Revealing the new system through a post about a related feature on the Twitch Support channel rather than a conventional announcement on the Twitch Blog is hardly a high-visibility way to get the word out, and a response to a more recent post by Bussey about the feature muddied the waters by saying "Twitch is not training GenAI models on streamer content."

But Twitch's own account settings FAQ indicates that's exactly what it's doing. Under the question "What are 'generative AI content models,' and what does it mean for my content to be used to train them?" Twitch wrote:

"Generative AI content models are AI tools that can help create or synthesize content outputs like text or images when given instructions or prompts because they are also trained on content. When you allow training, that means that your content may be used for future Gen AI model improvements. An example of what happens when you allow your content to be used for training a GenAI content model is that your audio might help refine models that create speech to text, which would help improve captions at Twitch but would also help improve captions across Amazon."

That seems clear enough, so what does that statement on X mean? My best guess is that Twitch is trying to claim that Amazon is using streamer content to train generative AI models, not Twitch itself—a distinction completely without a difference given that Twitch is part of Amazon. I've reached out to Twitch to ask for clarification.

The FAQ also notes that opting out of generative AI training "does not opt you out of all AI or machine learning uses at Twitch." The platform can still use your data, including streams, VODs, and all the rest, for "AI-supported features" like auto-generated captions, although that is not retained to train models that can generate new content.

This is hardly Twitch's first big botch. If anything, the platform has been a hotbed of bad decisions over the years, on topics ranging from sexual content to Donald Trump, Palestine, "politics and sensitive social issues," and when you can and can't say "Zionist"—and that's just within the past three years.

But this move could be more impactful than most: Mike Futter, co-founder and analyst at game consultancy F-Squared, said Twitch's new generative AI training system has potentially far deeper consequences for streamers than might be immediately apparent: "What happens if a creator streams your game and is opted in? Does Twitch think it has the right to feed a developer's game into its AI slop factory?"

What happens if a creator streams your game and is opted in? Does Twitch think it has the right to feed a developer's game into its AI slop factory? This has deeper ramifications than it seems on the surface.

— @futterish.fsquared.biz (@futterish.fsquared.biz.bsky.social) 2026-08-12T22:17:07.463Z

"I think a lot of studios are going to be talking to their lawyers over the next couple of days," Futter told PC Gamer. "For those of us who have avoided AI like the plague, this is a clear and present danger to our businesses and our creative work.

"In a period where our industry needs as much support as possible, especially from the companies that use our content every day to enrich themselves, this reads as extremely hostile."

"We're a staunchly anti-AI studio, and this now creates a real threat to our creative work," added Futter, who's also director of operations and publishing at indie developer Causeway Studios. And while he thinks Twitch and Amazon will face some pushback from game makers over its gen AI training plan, "this is David and Goliath. We're going to need a critical mass of studios and creators to be loud about this."

Given Twitch's history of switchbacks, I wouldn't be surprised to see this new feature dropped because of the backlash—but given Amazon's determined embrace of generative AI, I also would be surprised by it. Either way, the bottom line at this point is that the new generative AI training model is a mess for Twitch on multiple levels, and it seems more likely to get worse than to go away anytime soon.

We've been hearing for quite some time that Microsoft is planning on making changes to Windows 11 that focus on performance, quality-of-life, and just actually good things. Rather than, y'know, AI gubbins almost nobody was asking for.

And now, according to Windows Latest, the most recent update for Windows 11 includes a feature that's supposed to make apps launch quicker.

The 'low latency profile' launched earlier in the year but it now apparently covers apps in addition to Windows elements such as the Start menu. This feature essentially allows the scheduler to have the CPU boost up to fast speeds and drop down again quicker, which is good for short burst tasks like opening apps.

You should get this feature automatically with the latest Windows 11 KB5121003, build 26200.9168 / 26100.9168 update, and you can check whether it's working by keeping an eye on your CPU clock speed and seeing whether it boosts up near max in software like HWInfo.

This isn't the only useful update we've seen for Windows 11 in recent weeks. Although initially just for Insiders, recent improvements to the Start menu look promising. And while a recent update did seem to mess up my monitor (or, perhaps more likely, my AMD driver and how it interacts with my monitor), that issue has disappeared now, so either Microsoft or AMD presumably fixed it on the down-low. It wasn't a problem for too long, just a few days.

All this being said, there is certainly a slight taste of 'it's about time' regarding these actually useful Windows 11 improvements, especially since Microsoft is reportedly charging OEMs significantly more for Windows licenses now. At least we should be getting a snappier, more performant experience for that extra cash. Keep these updates rollin', Microsoft. I'm here for them.

This week, Nvidia announced it will be receiving "over $500 billion of third-party capital" from banks and investment companies. And now, in a blog post explaining how 'AI factory' compute is "becoming an investable asset class", Nvidia has answered the question, "Is this circular financing?"

The company says: "This initiative is designed to address that concern. We are bringing independent, long-term institutional capital into the AI infrastructure market.

"The demand is real: it comes from frontier AI labs, AI-native startups, enterprises, cloud providers and countries building AI services. The capital providers independently underwrite each project — including the customer, demand, utilization, cash flow and residual value. NVIDIA provides the platform; the investors make independent financing decisions.

"This is the beginning of an open capital market for AI infrastructure."

BlackRock CEO Larry Fink reportedly said, "We need to raise this money as fast as possible... This is going to be creating a huge amount of jobs."

Let me get the positivity out of the way with first: Yes, it's good that Nvidia is addressing the concern that the AI market finances itself circularly, by having AI companies fund other AI companies and vice versa with the same pot of money going around and around. Alright, that's the positive sorted—on to the negative.

First, of course Nvidia can say so now, after months of circular financing has already occurred. Now that it's opening up to widescale financing beyond the AI industry itself, it must be a lot easier for it to talk about. But where was this recognition six months ago?

Second, in the world of bigwig investment, it's hard to ever fully distance yourself from circularity. BlackRock, for instance, alongside Microsoft and Nvidia, has previously bought almost 80 AI facilities in a $40 billion deal. So I doubt funds already wrapped up in AI investments will balk too much at getting Nvidia a bunch of money for more AI building, as some of that will presumably trickle back in their direction, too.

Third, it's kind of unsurprising that big hedge funds like BlackRock would want to keep the AI hype going given the economy itself is arguably now somewhat dependent on it. Most economic analysis seems to agree that the AI industry is indeed a (giant) bubble, and that when it pops it could spell very bad news for the economy in general.

To wit, New York Times financial columnist Andrew Ross Sorkin says of this $500 billion partnership: "Mark Monday on your calendar. It’s a date we may look back on years from now as either the inflection point in the AI boom—or the moment it got so leveraged that a crisis began to form."

Finally, while it might be good that Nvidia has acknowledged concerns about the AI industry's financial circularity, it's far from clear that a $500 billion investment from Wall Street adds more solidity to the market. We have to remember that these investments are made on the promise of AI actually returning on investment, something we've seen little evidence of so far.

To my eyes, it looks a lot like tech CEOs gritting their teeth and crossing their fingers that the bubble doesn't pop before AI makes some miraculous, superintelligent breakthrough that grants an exponential ROI. And if that's the case, this $500 billion investment might mean more of an "open capital market" for AI, but it could still just be buying time before the inevitable *pop*.

It might feel like an eternity that we've been living in this AI era, but really we're only at its advent. As such, many of the risks associated with it have until now been merely hypothetical. However, we're now seeing research filter out from security and software research teams that show some genuinely concerning behaviour, in particular when it comes to AI agents.

On the one hand, we have Island Technology explaining how it's uncovered thousands of malicious GitHub repos disguised as AI agent skills and Model Context Protocol (MCP) servers that the agents are at risk of downloading of their own discretion. On the other hand, we have the AI Security Institute (AISI) showing how the (unrestricted) AI agents it was using for cybersecurity attempted to dupe real people, using fake identities and pressuring people into accepting malicious code.

I remember when OpenClaw ('Moltbot', back then) first entered public consciousness with promises of actual agential behaviour to help solopreneurs and the likes do, umm, stuff and things. Without said solopreneurs having to do said stuff and things themselves—just hook your bot up to your different apps and services, tell it what to do, and let it cook. That wasn't too long ago, either; I reported on it in January.

How time flies. Now, talking about autonomous, agential AI seems pretty normal. And we're starting to see what this actually means in terms of security risks. We'd seen cases of AI agents running amok before, of course, but these newly identified risks seem a little more concrete.

In the first case, Island Technologies has discovered that fake GitHub repos pose a real threat for AI agents. These kinds of attacks existed previously, of course, but they attempted to dupe real humans who could personally assess them and take responsibility for vetting things thoroughly before downloading. The difference is these repos are now dressing up as AI agent skills and MCP servers to specifically target AI agents, which could download these repositories of their own 'volition'.

The Enterprise Browser creator explains: "The most significant shift is a technique we call AgentBaiting. An AI agent searching for a new capability such as a Skill or an MCP server can discover a campaign repository on its own, treat the attacker's Readme as legitimate documentation, and hand the installation instructions to the user.

In our testing, Claude Code, Gemini, and ChatGPT all surfaced malicious campaign repositories without ever being shown a link. A playbook built to deceive people now deceives the agents acting on their behalf."

In at least one run, while Claude didn't download the relevant malicious repo in the company's testing, it did recommend it as a backup. In other runs it did detect malicious code and refused to recommend it.

As if this wasn't bad enough, there's the potential that AI agents could some day be the ones that create such repos, at least if a recent incident disclosed by AISI is anything to go by.

According to the institute, in some of the cybersecurity challenge runs it tasked AI agents with, "an AI agent took autonomous, unsanctioned action on the live internet, targeting real people and organisations."

More specifically, in one case, AISI says, "An agent tried to insert malicious code into an open-source project. In an attempt to get the code approved, the agent engaged in social engineering — creating fake online identities and using them to pressure the project's maintainer to approve the code. A human maintainer caught and refused to approve the malicious code."

"The agent researched the project's human maintainers, created multiple fake identities, and used the fake identities to socially engineer a real maintainer into approving the code. When the agent’s pull request was challenged in public, it edited its earlier activity to appear harmless and considered adopting a fresh identity to continue. The agent used Tor to bypass some network restrictions on GitHub, which is what first triggered AISI’s security alert."

"As part of the same effort, the agent tried to contact real people directly, sending messages and files through an online file-transfer service to persuade them, or their own AI coding tools, to run malicious code. Some messages carried harmful payloads, and some were attempts at social engineering; targeted at real people—something we’ve never previously observed."

The important caveat to note here is that these AI agents had had their reins taken off and were permitted wide-spanning internet access. The usual preventative measures built into these agents in public models had been disabled.

That being said, it's hard to imagine that really serious malicious actors wouldn't be able to do the same and remove those safeguards. In which case, I can't help but wonder if there's a chance that in the future we'll see AI agent-designed or agent-injected malicious repos being unwittingly accessed by AI agents that don't detect it.

There are, of course, always risks associated with new technologies, so I don't want to stoke fear too dramatically. But it's hard not to worry about such research when it involves a technology that is fundamentally designed to have some level of autonomy. That's the key issue here, and in both cases—whether acting maliciously or unwittingly downloading malicious software—it's this agency that is the primary issue. Let's just hope improvements to safeguards keep up with improvements to AI agent capabilities.

Saber Interactive's upcoming Rideshare Stimulator drew some eyeballs today when Saber CEO Matthew Karch confirmed that its trailer includes a hidden jab at Gearbox boss Randy Pitchford. But that same trailer sparked a grimmer reaction from former lead writer Stella Sacco, who said on Bluesky that she was replaced on the project by ChatGPT. Saber, however, denies the claim, saying that no writers were replaced with AI "for Rideshare or any other game."

"I didn’t even see this had been announced," Sacco wrote earlier today. "I was lead writer on this one! And Saber replaced me with ChatGPT midway through development. All the passenger voices were AI too. Either they changed direction at some point or they’re not disclosing it on Steam."

I didn’t even see this had been announced. I was lead writer on this one! And Saber replaced me with ChatGPT midway through development. All the passenger voices were AI too. Either they changed direction at some point or they’re not disclosing it on Steam

— @antlervel.vet (@antlervel.vet.bsky.social) 2026-08-13T01:48:31.362Z

Sacco also shared an image of a December 2023 social media post, in which she had written, "I got replaced on one of my games at work. By ChatGPT." Sacco's LinkedIn account indicates that she was lead writer at Saber Interactive from July 2023 to July 2024, during which she worked on Stuntman Hollywood and other games.

In a subsequent post, Sacco said the decision to replace her with a chatbot was made by "one to two AI dummy higher-ups at Saber," and that she has no issues with Rideshare "Stimulator" developer Unigine. "My boss, to his credit, argued over and over with the c-level who pushed the AI—'we already have Stella and she’s funny and she’s good at this, she’s done live games before so live ops are no problem,' etc—but the suit insisted 'gamers love new tech' so using AI would be 'great marketing'," she wrote.

The allegation resulted in immediate blowback on YouTube and Steam, over both the alleged use of AI and the lack of disclosure on the game's Steam page. In a statement provided to PC Gamer, however, Saber denied Sacco's claim, although it acknowledged the presence of AI within the game.

"Neither Saber nor Unigine have replaced any writers with AI for Rideshare or any other game," the studio said. "In fact, Saber's staff of writers has tripled since 2023. Rideshare's story was written entirely by real people.

"The game will also feature an experimental, free-ride mode that by necessity incorporates AI as the number of passengers in the game is infinite. We are firm believers in using AI tools to improve or enhance gameplay experiences and improve NPC behavior as we have done since Saber was founded in 2001. As the industry has continued to contract, Saber continues to hire. None of our 3000+ employees have to worry about being replaced by AI.”

In response to Saber's statement, Sacco told PC Gamer, "They can say what they want—I was there, I even posted on Facebook about it at the time, and I have mentioned it on Bluesky several times without using the game’s title.

"They took me off the project to use ChatGPT to write the rest of the text, including planned live updates. If they changed plans in the intervening years, that’s great, but again: I was there, and it was my job to write the game, and then it wasn’t once Matt Karch decided using ChatGPT would be great publicity for the game."

The Rideshare "Stimulator" Steam page says "each ride presents players with unique and unpredictable passenger stories, ensuring a diverse and engaging gameplay experience." That's not the most compelling bit of marketing I've ever read (and doesn't that make you wonder?) but it does imply the importance of strong narrative and writing to the overall experience. I had a friend who drove taxi years ago and he wasn't a huge fan of the work but he loved the people, and I think that will hold true in Rideshare "Stimulator" too: The mechanics of driving will provide a skeleton, but it's the people in the game world that will give it life—or not, as the case may be.

At the same time, though, it also jibes with what Saber said about the game's "free-ride mode," which essentially promises the opposite of strong narrative and writing. Specifying that "the story" was written by humans, as opposed to everything, is also effectively meaningless because it could refer to anything from the basic plot to a specific script—and excludes narrative aspects like dialog, which is what ultimately makes a game like this work.

In spite of the admission, the Rideshare "Stimulator" Steam page is also notably lacking an AI disclosure. In a follow-up statement, Saber told PC Gamer "the Steam page will be updated once the feature set is finalized. We are touting this experimental mode as a feature, not shying away from it."

I've spent a large part of my morning reading the words of Mark Zuckerberg. 6,434 of them, to be precise, if we're including the "Mark" signoff at the bottom. The Meta CEO (and everyone's favourite smart glasses advocate) has written what can only be described as a manifesto regarding upcoming "superintelligent" AI technology—and I'll be honest, my eyes have glazed over more than once.

The piece/blog post/rambling diatribe has been posted on Meta's website, under the title "The Future is for Everyone: The Path to a Positive AI Future." If you'll allow me, I'd like to quote some choice insights:

"The defining questions of our age are who will have access to superintelligence and what will we direct it towards. Will it be centralized and restricted to a few institutions, or will it be a tool that empowers everyone?" asks Zuckerberg. "We propose a philosophy based on individual empowerment as the source of prosperity, invention as the primary purpose of superintelligence, and balance of power as the foundation of safety."

How will that work out, exactly? Well hold on for multiple paragraphs, because we'll get there eventually. Before that, though, Zuckerberg has some scene setting to do.

"Novel ideas and major steps forward rarely originate from established institutions alone. They came from the brothers in a bicycle shop who believed people could fly, the bookbinder's apprentice with no schooling who figured out how to generate electricity, and the kid in a garage who thought personal computers could be for everyone," espouses the Ray-Ban-ed one.

"We believe this will continue to be true. As everyone gains more powerful tools, each person will become more capable of shaping the future, not less."

Sure, we'll all be like the Wright Brothers or Michael Faraday. Got it. And superintelligence will be very empowering, and innovative, and we'll all want it. How do you plan to put it in our hands, Zuck?

"Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it. This has the potential to begin a new era of personal empowerment where individuals can use this powerful new capability to reach their full potential, pursue their interests, and improve their lives and the world more than ever before."

Same thing, but sure. Wait, though, because the Meta CEO has five starry-eyed predictions of how this'll all play out. They are:

1. "Everyone will have incredible tools for creation to express your ideas"

Here, Zuckerberg uses the example of his eight-year-old daughter, who is apparently already coding thanks to AI. Now they're designing a robot together. Lovely. The good news for the rest of us, though, is that "everyone will soon have invention superpowers." Can't wait.

2. "Everyone will have powerful tools to create new businesses"

This seems a lot like the expression and invention thing, but applied to the serious business of, err, business, and all the "ideas" that will now be easier and more feasible to implement thanks to superintelligent AI. "I predict that this will not only lead to much greater economic growth, but also more employment over time rather than less," says Zuck. He's not alone there, at the very least.

3. "Everyone will have a personalized tutor and coach with a PhD in every subject"

Because superintelligent AI will know all, presumably. It definitely won't be like current AI, which gets things wrong all the time. That's the old stuff. The new stuff around the corner is much better, promise. You'll be able to trust your education to it, even. I cannot see any flaws with this idea.

4. "Everyone will benefit from scientific advances and be able to contribute to scientific progress"

This one's about AI biotech advancements, and it gets real scary, real fast. Observe: "Since there is a long tail of health conditions and everyone's biology is different, people's direct involvement will help unlock personalized therapies and expedite trials to match the pace of invention and enable industry to move beyond focusing disproportionately on common diseases."

On paper, AI actually shows some promise in helping with medical research. I'm not entirely sold on the idea of handing over my medical data to AI developed by companies that have had some, shall we say, controversial data privacy issues in the past, though. Just a me thing?

5. Everyone will have free or affordable access to these tools

And finally, Zuck says that there'll be free versions of superintelligence that will be accessible to "billions of people," but for those who "want to pay to use more compute," there will be a dynamic auction mechanism that will "guarantee that everyone gets the lowest price possible... while also ensuring the capacity is used for whatever people collectively find most valuable."

The AI future will be created by all of us, you see. Not just massive corporate interests like Meta, or OpenAI, or any of those other companies you read about every week getting up to all sorts of dubious AI shenanigans. The power should be in the hands of the people. As we make it, when we decide. Don't you feel better already?

Wait, though, because there's more. "People and institutions with competing interests naturally check and balance each other to lead towards positive outcomes," says Zuckerberg. Therefore, if everyone's got access to superintelligence in some form or fashion, it'll all work out in the end.

I'm no economist, but that sounds awfully like "the free market will even itself out eventually." Which hasn't exactly worked out in practice, as far as many of us are concerned.

What Zuckerberg seems to be getting at, in among all the starry-eyed platitudes regarding our bright AI future, is that superintelligence should not be in the hands of just governments, businesses, and small numbers of individuals.

We're this close to a bright, kumbaya-singing, superintelligent AI future—as long as Meta keeps being the wonderful company it already is and Zuck's principles become sacrosanct. And we all buy the rhetoric, of course. Meta is helping to build the democratic AI future, so remember who you support in the dark times to come.

But what about jobs? Good news, everyone. It'll all be fine:

"People fear that automation will outpace individuals' capability growth, leading to job displacement followed by a difficult period as people learn new jobs," the rambling continues. "But there is no rule that AI must increase automation faster than it increases individuals' capabilities or demand for new skills.

"Recent statistics suggest it may be more likely that individuals' capability growth could match or outpace automation, in which case people will gain the ability to do many new things before their current jobs change. This would lead to a healthy balance and potentially even job growth."

Which flies in the face of the predictions of the head of the IMF, the AI-related layoffs at Amazon, and a number of other companies that are shrinking their workforces already as a result of AI implementation. It'll all even out, though, because those laid-off employees would have become data center specialists, or something really useful, if superintelligence had been around to educate them. Coming soon, we promise.

The good news, though, is that Zuckerberg recognises that it's not in "the world's best interests" for himself or his company to be the sole decision maker in how superintelligence is deployed.

This means that Meta's independent board of directors will review whether new models meet safety criteria, and Zuck encourages other companies to do the same. He also proposes for leading AI labs to provide training checkpoints for new advanced models (alongside technical staff) to governments, in order to help them gain security capabilities.

Zuckerberg is also in favour of more government policy around AI, through "close co-operation" between frontier labs and (presumably the US) government.

As long as America doesn't lose its AI edge, of course. "It is also important for national security that the American AI industry continues to lead," says Zuckerberg. "We should accelerate the ability to build infrastructure—whether that's energy, data centers, or silicon—and build trust and alignment with communities across the country."

Stop protesting those data centers, folks. America needs to stay ahead, but we'll happily meet with you and tell you why. Methane turbines near your house? Nah, that's the old AI. In the future, it'll all be about working with the local community. Promise.

I could go on, but Zuckerberg's ramblings have taken up enough of my day—and if you've got this far, yours as well. Basically, superintelligence is going to be great, and Meta is working to ensure we'll all get a slice. Your jobs are fine (or will change, but will still be fine) AI companies are going to protect you from tyranny by building a democratic superintelligent AI future, and we can all sleep soundly in our beds at night.

Personally? I don't feel entirely reassured.

Asus Armoury Crate is not my favourite bit of software. It looks all high-tech on the surface, but any time I use it for more than five minutes, I'm reminded of the fact it's rather messy. It's also hit the headlines previously for patched security vulnerability issues—and good news, folks, here we go again.

Per the Asus security advisory board, Armoury Crate, Asus GPU Tweak III, and GPU Tweak II should all be updated to ensure "optimal protection," after the discovery of a new security vulnerability.

Listed as CVE-2026-8917, the issue has been given a score of 8.4 out of 10, with a "high" severity rating. Which is, funnily enough, the exact same rating as the previous issue we wrote about last year. "An IOCTLY vulnerability allows a local attacker to write a specific value to an arbitrary memory address, potentially leading to privilege escalation," says the posting.

Well, that sounds like something that should be patched up at the earliest possible moment. The CVE listing doesn't mention Armoury Crate specifically, but the Asus page mentions that "a component used" by the software is part of the update, so time to get installing.

GPU Tweak III, II, and AI Suite 3 will need the latest versions installing from their respective support pages, whereas Armoury Crate can be updated from within the app itself.

As an aside, Asus isn't the only company that patches out security vulnerabilities for its software on a regular basis, so if you've got something hanging around in your system tray from another hardware provider (particularly system/mobo software that came with your machine), it's a good idea to check for updates on a regular basis.

It's a dangerous world out there, and new security holes are patched all the time. Get thee to the update menu, and keep your ear to the ground.

Taiwanese news site United Daily News reports that, according to one source, Microsoft is charging Original Equipment Manufacturers (OEMs) significantly more for Windows licensing fees this year than in previous years (via Wccftech). These are the fees that PC makers, for instance, must pay to Microsoft to package their machines with the operating system.

UDN says (via machine translation): "A PC brand executive stated that Microsoft increases licensing fees for Windows OEMs every year. In the past, licensing fees mostly increased by single digits each year, but starting in July this year, the average increase has reached 7% to 10%, which is a significant increase compared to previous years."

I can't confirm whether UDN's source is correct, but given Framework recently said it's seen an increase in Windows license costs, I have no reason to doubt it.

Apparently Microsoft charges OEMs more for Windows licenses if they're used with higher-end CPUs—which seems like a strange decision to me given it shouldn't actually change anything about the operating system being delivered, but I'm not a business bigwig so what do I know?

I don't need to tell anyone that things are already expensive right now, and not looking to get any better any time soon. That's primarily on the physical hardware side of things, though, thanks to the global memory shortage. In fact, I recently did the math and discovered just how bad of a picture is painted by comparing RAM prices from 2007 to today. And some who are in the know predict memory supply might be constrained "even beyond 2030."

We'd not heard much in the way of software prices rising, though. Software—and some specific kinds of hardware such as monitors—have been somewhat of a last bastion for pre-AI-induced-shortage prices. Alas, it seems Microsoft might not be content allowing memory companies to have all the fun.

7–10% might not seem a lot, but it all compounds on top of an already expensive market. OEMs are unlikely to eat the costs, so eventually such price hikes could hit end-consumers like us. And while we might be able to select OS-less systems at checkout, OEMs could increase prices across the board regardless if their margins keep getting slimmer.

Microsoft in general doesn't seem to be struggling, either. Despite its Xbox division performing worse than last year, the company's overall revenue is actually up by 18%, comparing the previous quarter to the same quarter last year.

That being said, the company has also lost exclusivity with OpenAI, meaning the AI company can now use cloud hosting other than Microsoft Azure, after a few years of being restricted only to those servers for training and hosting. That close partnership was part of what ushered in the 'Copilot' AI PC era. The partnership still exists, but OpenAI isn't as closely tied to Microsoft.

Still, Microsoft seems to be doing just fine, so it's difficult not to be a little frustrated that the company is potentially adding on to the cost of PCs. Is that someone saying 'the year of Linux' I hear?

Earlier this year, Google Chrome annoyed a whole lot of adblock users by killing support for Manifest V2 extensions. It's a Manifest V3 world now, which means the permissions and APIs that thoroughgoing adblockers like uBlock Origin rely on are a thing of the past on Chrome (it's since been replaced by a pared-down version: uBlock Origin Lite). Now, Microsoft Edge is following suit: "Over the next few months, MV2 extensions will be gradually turned off by default," says Redmond.

All of which is red meat for the folks over at Mozilla Firefox, the free-and-open-source browser that once reigned supreme on our PCs before Chrome came and ate its lunch. In posts across social media, Firefox kicked Edge (and Chrome) in their self-inflicted bruises, declaring that "Our support for uBlock Origin isn't going anywhere", a post that came days after it wrote "Reminder that you can still just… use ad blockers on Firefox. Wild concept, we know."

Now, like I said, it's not like Chromium-based browsers are murdering adblockers wholesale. If you wanna block ads, you can still use less-powerful extensions to do so, but it's a simple fact that if you want a full-fat adblocker in the uBlock Origin mould, the megacorps are rapidly shutting down ways to do that.

It's also true that Firefox isn't your only option—there are several other browsers out there still touting full-fat adblocker support—but it's probably the biggest name out there fighting the good fight.

Our support for uBlock Origin isn't going anywhere.

— @firefox.com (@firefox.com.bsky.social) 2026-08-11T10:55:07.466Z

Anyway, Firefox's braggadocio prompted an obvious question: hey gang, when's this vaunted adblock support gonna make its way to iOS devices? Alas, there's no hard timeline there, but Mozilla did say that it's actively working on making adblock work on Firefox on your iThings.

In fairness, Apple's strict rules mean that Firefox on iOS (along with other browsers) are more like reskins of Safari than they are full implementations of the desktop browser on mobile. But hey, if people are using Firefox in 2026, it's likely in part down to its robust extension support. Better to be able to offer some form of it than not.

Hot on the heels of this Codex Micro/OpenAI mini keeb for AI agent monitoring is this... thing that Keychron seems to be experimenting with.

Apparently it's a tiny, three-button, wireless mini keeb with a scroll wheel, but hidden inside is a microphone that lets you talk to your AI agent as it vibe codes. Yes, we've gone too far as a species. I entirely agree.

"Imagine talking to your coding agent without switching windows," says Keychron. Ah yes, the thing we've all been dreaming of but haven't yet been able to obtain. I'll take three, even if they do look like the tentacles from, err, Day of the Tentacle.

Alright, I'll dial back the snark for a second. For a start, it looks like Keychron is simply polling for feedback on the idea—and the X thread is full of people suggesting features, so there's probably a genuine market.

And second, while I'm somewhat salty about the rise of sloppy AI-coded software, vibe coding can produce some genuinely useful results. I'm not sure how voice control meaningfully adds to the experience (beyond swearing uncontrollably at the AI if it deletes your database in a single swoop), but hey, it's not really my field.

The top comment suggests adding vibration, which would "10x your total addressable market." I'm not sure what they mean by that (nothing to do with the shape, surely?), but Keychron responded: "Dammit, I never even thought about motors."

So, watch this space for buzzy-buzzy fun times. My favourite reply, however, suggests the addition of a panic button, to which Keychron responds: "An instant STOP WHAT YOU'RE DOING button?" Yes, that would seem prudent. No matter which way I look at it, really.

ChatTJB, according to its Crunchbase profile, is a service that "provides a personal conversational response where an individual directly receives, processes, and writes answers to user inquiries". That's technically true, but not in the way you think… because it's just sorta some guy.

Futurism got to sit down with Tucker Bryant, the man behind ChatTJB, and as you might be able to expect, he admits, "I tend to overcommit to bits."

If you're wondering how far he's willing to commit to the bit, Bryant rented out a billboard for $6,000, which says ChatTJB is the "leading chat interface powered by AI". AI in this case stands for 'Average individual".

He has reportedly been personally writing responses to users over 1000 times a day since starting the project, answering questions on what to eat for dinner, what colour to paint their homes, and he has even given relationship advice.

As Futurism notes, "Like a chatbot, he can also generate images upon request— in the form of graphics that he hastily sketches, of course."

Bryant is a Stanford graduate and a former Google employee who claims the idea came off the back of his partner telling him about "cognitive surrender". Effectively, it's argued that those who use AI chatbots are likely to blindly trust the information given to them, even when the information is incorrect.

Bryant says, "We often, or increasingly, trust LLM outputs so easily that we stop bringing our own creativity or critical thinking into the responses we receive, and over time that weakens our ability to be fully-functioning, sovereign human beings in the world."

Though Bryant was expecting some trolls and acknowledges ChatTJB is a bit of a troll itself, he says it has been "really sweet" to chat with real people and even noticed some users asking more serious questions. One user reportedly said, "I’m on the first night of my honeymoon and I don’t feel relaxed, is that okay?" and Bryant "switched out of super-casual-mode into supportive-friend-mode.”

Though that does raise some concerns. If users are talking to ChatTJB and are under the impression they are talking to an AI anonymously, they could reveal personal information in the process to another human being. That could become more of a concern as ChatTJB has seen interest from users online who want to become the bot on ChatTJB.

However, Bryant is taking it slowly and figuring out how to get volunteers involved. "This was just intended to be an art project. I think it’s beautiful that people want to do something less ridiculous with it than what I had planned."

He continues, "I also don’t want people to get hurt, so I’d rather people wait a long time for their prompts than open the floodgates to anyone and everyone. So, a long preamble to saying: I’m starting to let some approved volunteers try answering questions while testing out some moderation tools."

Bryant argues that ChatTJB is a satire of AI and the culture surrounding AI advertising. “That’s the highbrow answer. But the lowbrow answer is just as authentic, which is that I also thought it would be funny."

Bryant is not explicitly anti-AI, though. He says on his Instagram, "AI can do some really cool things, but sometimes all you really need is to talk with an average individual."

If one thing is true about AI, it's that there's a lot of money in it. Nvidia became the first $5 trillion company because of its role in the explosion of AI data centers, and OpenAI has been rounding up tens of billions in investments. So you'd think AI companies could buy pretty much whatever they want—but one farming family in Northern Kentucky showed that's not the case by turning down a $26 million bid.

As reported by WKRC-TV Local 12, Ida Huddleston was offered the sum for half of her 1,200-acre farmland this year, which an unnamed company would turn into an AI data center. Huddleston's daughter, Delsia Bare, reportedly said she would "stay and hold and feed a nation" and that "$26 million doesn't mean anything."

AI data centers have become an ethical and ideological battleground, with Elon Musk's SpaceXAI only this week agreeing to take down 69 temporary methane turbines in Memphis over the next year in reaction to resident concerns over local disruption.

In this case, Bare claims there's history to the site that the data center company would be ignoring. She told Local 12, “My grandfather and great-grandfather and a whole bunch of family have all lived here for years, paid taxes on it, fed a nation off of it." She continues, “[They] even raised wheat through the Depression and kept bread lines up in the United States of America when people didn’t have anything else.”

Bare claims to be among "dozens of landowners" approached by the anonymous buyer, who reportedly represents a major AI company. Notably, Local 12 claims that the land where Bare lives is worth around $6,000 per acre, which makes the offer around ten times the actual value.

Huddleston tells Local 12, “They call us old stupid farmers, you know, but we’re not. We know whenever our food is disappearing, our lands are disappearing, and we don’t have any water—and that poison. Well, we know we’ve had it.”

Huddleston is also sceptical of claims of economic growth as a result of the AI data center. “I say they’re a liar, and the truth isn’t in them. That’s what I say. It’s a scam.”

Data center pushback has been a common theme of the growth of AI, with concerns of both an environmental and ethical nature being front and centre. And AI companies have a lot of money to throw around.

Even if you own a lot of land, $26 million must be a life-changing amount of money to anyone but the richest of the rich; it would even be enough to buy some of the RAM that AI data centers have been gleefully sucking up over the last few years. Just a thought.

PC gamers (and PC users as a whole) find themselves in an increasingly expensive landscape, in part due to the AI data centre industry's apparently bottomless appetite for memory. And yet, AMD CEO Lisa Su reckons "AI PCs will become more important" for the average consumer.

Lisa Su told investors during AMD's most recent earnings call, "As we go into 2027, I think we have a strong product portfolio that is coming on board to address not just the traditional notebook and desktop markets, but also as you think about a more AI-centric PC experience, like what we have been talking about with the Ryzen AI Halo."

To be fair to Su, at the start of the year we did describe the Ryzen AI Max+ 388 as a chip that looks perfect for the ultimate in handheld PC gaming. For more context, AMD has just reported record-breaking total revenue of $11.5 billion in Q2 alone. A decent chunk of that is down to demand from data centres. Even so, AI PCs feel like an odd play when consumers are staring down increasingly runaway hardware costs.

Su seemed undaunted speaking to investors, saying, "I remain optimistic about the PC market as an important way for us to reach a broader set of users and our ability to grow ahead of the market."

For a little more context, the gaming section of AMD's business has shrunk by 31% in part due to high hardware costs (though the company likely saw that coming). It's perhaps less surprising then that AMD has been selling a lot of lower-end and mobile Ryzen processors this past quarter, rather than the newest gamer-geared GPUs and CPUs.

Su reflected on the current volatility of hardware pricing, saying, "The market itself will depend a bit on some of the components. We’re all watching the component costs and how that will play out over the next couple of quarters."

The client segment of AMD's business, "which primarily includes CPUs, APUs, chipsets for desktops and notebooks," made around $3.1 billion outside of gaming hardware (the client and gaming parts of the business "do not qualify as [...] separate reportable operating" segments, though AMD does still report their financials separately).

As such, Su told investors that the client business's "first half performance has been very strong." Though AMD expects the market to see a decline in the second half of 2026, Su went on to say she expects to see year-on-year growth in this more consumer-focussed sector. Even if it doesn't meet these expectations, data centre demand will likely more than buoy the company.

Su said, "I think our portfolio and our rising content in enterprise are very positive for our client mix."

That's all very well and good, but if big tech continues with its AI infrastructure build-out, that's just going to continue to drive up the price of memory and more besides. If AMD truly believes NPU-equipped AI PCs are the future, who does it think will be able to afford them?

Nvidia hasn't been just a gaming-focussed company for a fair few years now, dining well off of the booming AI industry. AMD, one of its main rivals, is much the same, with the company feasting on the data centre segment in particular this past quarter.

In total, AMD reported record-breaking revenue of $11.5 billion in Q2 alone. Demand from data centres made up $6.7 billion of that, doubling what AMD made in the sector last year. So far, not so surprising as Q2 2026 financial results go—especially as gaming revenue apparently saw a year-over-year dip of 31%.

Instead of, say, GPUs or CPUs for gaming PCs, AMD has been selling way more lower-end and mobile Ryzen processors, with its client business raking in $3.1 billion as a result. That's a 23% increase compared to Q2 2025, with AMD attributing the uptick to "a 29% increase in unit shipments of client processors, partially offset by a 3% decrease in average selling price of client processors." Lower-spec processors generally have a lower asking price, after all.

In comparison, though, AMD's gaming segment only made $779 million in revenue. Gaming's dip is no surprise for a few reasons. For instance, a custom AMD SoC sits at the heart of the PS5; the console's age combined with how much its price has risen recently means that Sony is shifting fewer funky white boxes, and AMD is selling fewer custom chips as a result. Same goes for the Xbox.

To speak more broadly, the memory supply crisis has not been kind to PC gamers either. For those who have apparently been secluded under a cool mossy rock for the last year or so, AI data centres have been hoovering up all of the RAM and SSDs, driving up prices not just for memory but plenty of other consumer-geared components as a result. With high prices on consumer hardware—like AMD's higher-end processors and graphics cards for gaming—there's going to be much less demand.

Hence, the gaming part of AMD's business has shrunk, though the company continues to grow thanks to the continuing AI infrastructure build-out. Not only did AMD report record revenue for the second quarter of 2026, but it also sailed past Wall Street's no less eye-watering estimate of $11.25 billion.

In short, AMD isn't just a gaming-focussed company, and clearly hasn't been for a fair few years now.

Though I write a fair amount about AI agents, I do not use them. This is because agentic AI applications often introduce a more than medium-sized headache in terms of cybersecurity. For instance, one security research recently explained how a Word Doc could be leveraged to get Copilot to spread an AI worm.

AI researcher Håkon Måløy breaks down how the attack in a recent blog post, writing, "An attacker places hidden instructions in a document that is later used as source material in Copilot for Word. Copilot may interpret those instructions as part of the user’s request, causing it to manipulate the document being drafted or edited. Copilot may then also copy the hidden instructions into the resulting document, turning that document into a new carrier."

The attack involves a "JSON-formatted malicious prompt." From this prompt injection of JSON-formatted data, a chain reaction wriggles into action. The worm replicating through 'carrier' documents scooped up in further Copilot-assisted workflows—the original document that kicked off the whole thing doesn't even need to be present. Worse still, the attacker needs no special access, they just need to "share a malicious document with the victim."

This isn't the first AI agent cybersecurity threat we've seen. For instance, back in February Meta's AI safety director recalled how she 'had to RUN to my Mac mini like I was defusing a bomb' when her OpenClaw AI decided to start deleting all of her emails. Last year, Replit's LLM-based coding assistant deleted a dev's entire database during a code freeze. But on the subject of Copilot PCs more broadly, cybersecurity experts warn that Windows Recall may be far from secure when it comes to protecting your personal information.

As for the Copilot AI worm, Måløy first disclosed the vulnerability to Microsoft back in March, though the attack is still reproducible at time of writing. Måløy offers a detailed disclosure timeline, and explains, "Two mitigation attempts, including a model upgrade, did not close [this] class [of vulnerability]."

As such, Måløy's blog post only broadly describes the type of attack possible, rather than going into detail about the hidden prompt that triggers the AI worm. The way this prompt is hidden doesn't require anything fancy, though, and may even be familiar to those who have ever wanted to catch someone out for using AI; at minimum, an attacker could format the malicious prompt as tiny white text on a white background.

"The prompt can be rendered as white text on a white background and in a small font size to conceal it from the victim," Måløy elaborates, "Since Copilot for Word strips all text formatting like color and font size before passing the text into the underlying Large Language Model (LLM), this text remains fully readable to Copilot even though the victim cannot see it. The attack can be further concealed by embedding it in a seemingly benign document with task-relevant text."

Long story short, this is a fairly low effort, hard to trace attack without sufficient mitigation currently in place. Cybersecurity and antivirus company Malwarebytes offers a few tips on how to stay safe from this class of attack, including disabling Windows Copilot in Word or simply ditching the AI agent altogether.

Personally, I think Måløy's closing thought sums up the security risk of AI agents nicely: "Any system that integrates an LLM into a trusted workflow today must assume that attacker-controlled content entering the model’s context will result in compromise at some rate."

xAI has been at the centre of ethical and legal disputes over the last year due to the 69 mobile turbines it erected in Memphis to power AI data centers, and the environmental impacts of all that tech. Now, in accordance with an agreement with the Mississippi Department of Environmental Quality, SpaceX and xAI have agreed to take them down over the next 11 months.

In a blog post, xAI says that it will be taking down all 69 mobile turbines that currently power its facility in Southaven, Mississippi, and will instead be swapping to power from a 1.2 GW power plant it is currently constructing. Between August 2026 and July 2027, it is expected to have removed all turbines.

The blog post says, "We care deeply about being good neighbors and we regularly reconfigure power operations to reduce noise from the facility and minimize our impact on the local Southaven community." One could suggest that if xAI really cared about being good neighbours, it probably shouldn't have, as the SLEC argues, "operated with a stunning lack of transparency that left impacted communities in the dark."

xAI also says that it will make investments to equip SpaceXAI mobile turbines with emission control technologies going forward. The 1.2 GW plant will reportedly consist of 41 permanent turbines, all of which are authorised under a Clean Air Act permit.

Many AI companies have been desperately expanding data centers in the hopes of catching up with the competition. Meta just last year was reportedly expanding AI capabilities by housing data centers in tents, and is continuing to throw money at its AI infrastructure buildout.

The rapid expansion of AI has not only led to environmental concerns and companies building regardless of whether the infrastructure is there to handle it, but it has also led to the memory crisis. The reason you won't get RAM at a reasonable price these days is because of how quickly AI has expanded and how little supply there is to go around. It seems AI is unlikely to go under soon, so the memory supply will be unlikely to meet demand for at least a few years.

Even Samsung, one of the three biggest memory manufacturers, reckons next year will see a "severe" memory shortage that will last until at least 2029.

Still, even if it took kicking up a fuss and legal threats from the NAACP to do so, xAI complying with the Department of Environmental Quality is still a good move long-term. It just feels a little rich for SpaceXAI to pat itself on the back while doing so.

After two former Apple employees started working for OpenAI, the Cupertino-based company alleged that trade secrets went with them and straight to the creator of ChatGPT. After it asked for a preliminary injunction to stop those employees and OpenAI from accessing confidential information, OpenAI publicly announced its defence, and followed that up with a blog post for good measure.

Titled "Apple is getting this wrong", OpenAI argues that Apple has built a reputation for obsession over small details, yet "this careless, aggressive and oddly personal lawsuit sadly doesn’t live up to that reputation."

It notes that Apple claimed it contacted OpenAI in February, but that contact was actually given to the wrong person, as Apple's lawyers mixed two names up. It also notes that Apple claimed it talked to OpenAI's General Counsel, "which they now concede never happened."

The two former employees, named Chang Liu and Tang Tan, are both defended by OpenAI. OpenAI claims that Liu is accused of trying to access confidential information after leaving, though OpenAI shares text messages of Apple employees asking for Liu's help to locate files. OpenAI claims that the problem here is actually Apple failing to properly manage system access when people leave.

Apparently no such evidence exists for Tan, but OpenAI claims "Tang has always been clear with the team that we do not want, and must not use, any confidential information from other companies."

OpenAI claims that the preliminary injunction is "based on false information" and unnecessary, as it doesn't want Apple's trade secrets.

The blog post argues, "We wish Apple had raised these issues with us before they filed their lawsuit and would have been happy to clarify all of this with them. We have taken the allegations in the lawsuit seriously and offered to work with Apple to resolve this." It continues, "Instead, they’re trying to change their narrative, including by making vague accusations about other former employees, and will likely keep repeating this tactic."

Apple is yet to respond to OpenAI's public blog post, but the company is historically less likely to use the same tactic of publicly posting blog posts about ongoing court cases. In regard to Elon Musk's lawsuit against OpenAI, the AI company put up two separate blogs recounting its relationship with Elon Musk and Elon Musk's nonprofit claims.

Naturally, a blog post like this is intended to garner public support, which can then apply pressure to Apple. Apple will need to come back with a strong response, should it choose to give one publicly.

Remember the story about AI companies destroying millions of copies of books while scanning them to use as training data? In terms of bad optics, "destroying books" is right up there with kicking puppies, which is why the AI companies relied on third-party rebuyers to obfuscate what they were actually doing.

It didn't work, and now as 404 Media reports (having broken the original story), third-party book database company ISBNdb has deleted the part of their website advertising that they could be "your streamlined partner for sourcing printed books in bulk, tailored to your LLM training needs, delivered at the scale AI demands."

That page is still available on the Internet Archive, of course, as is their blog post about "Reframing the Destruction Narrative" in which they make wild claims like: "The book is not destroyed. Its value has migrated. The paper returns to the material cycle; the knowledge enters the intellectual one."

The idea that asking a chatbot to hallucinate a summary of a book has exactly the same value as actually reading it seems bugloving nuts to me, but what do I know? I'm not getting my knowledge from the intellectual cycle. Anyway, ISBNdb is now claiming it never actually offered the service it dedicated multiple parts of its website to, telling 404 Media, "The page was a test of market interest; no such service was ever brought to life."

Somebody out there sure is buying up a lot of old and obscure books, however. As reported by Guardian Australia, multiple secondhand booksellers have seen an uptick in online sales from third-party rebuyers, describing them as "waves of orders that did not fit with usual customer patterns."

Even though it's been ubiquitous for a number of years at this point, it still feels like generative AI is working on its sales pitch. OpenAI CEO and Elon Musk slapfighter Sam Altman took to X yesterday to offer a new "cool use case" for ChatGPT he recently overheard: "Connect your family calendars and explain your kids' interests."

"Every morning for the drive to school, have it make a podcast that talks about one kid's soccer game that afternoon, one kid's upcoming birthday, some news, etc." In other words, delegate the task of connecting with your blood relatives to a productivity tool. Gravity Falls creator Alex Hirsch has racked up over 120,000 likes for his reply: "What if you just talked to your children?"

When my dad drove me to elementary school, he would put on country music radio and I would try to explain to him why Spore was so great. We learned about each other that way. The thought of having those memories replaced by some LLM telling him "it's not just a little league game—it's an experience," well, it makes me think of that experiment from the '50s where monkeys had their mothers replaced by surrogates built from wire.

The response on X has been largely negative. Comedian Gianmarco Soresi quote-reposted Altman, saying "Cool use case of a vibrator I heard last night: you can go in separate rooms and not have sex." A quote-repost from Kathryn Jezer-Morton reads, "Surreal how stupid this idea is." Another comment from AM_Pines said, "Children are supposed to be educated with Creedence Clearwater Revival during the morning drive dammit, not some fuckass generated AI slopsicle."

Granted, your view of this might change if you believe, as Altman has said, that we're living through the singularity and AI has surpassed human intelligence. But that claim—and these musings that ChatGPT should help you talk to your kids and take care of your baby—accompanies a broader push by Altman, OpenAI, and other tech companies to resist regulation by leveraging the promise that AI will fundamentally change how the world works.

Original Story: First reported by Wired, AI company Anthropic revealed in a July 30 blog post that its AI agents have successfully hacked three unidentified companies since April. The whammy: Anthropic claims it only realized this after OpenAI's recent Hugging Face fiasco⁠—where one of its prototype agents hacked at least one other company⁠—led it to conduct a review of its own operations.

The incidents occurred as part of testing with an external firm, Irregular. The AI agents were supposed to be constrained to a simulation, hacking fictitious companies as part of a "capture-the-flag challenge." Notably, while OpenAI's alleged rogue AI incident occurred after the agent overcame its testing limitations, Anthropic stated that its models were mistakenly granted internet access due to a "misconfiguration" with Irregular.

"Neither we nor our evaluation partner were aware of this misconfiguration until we detected it through our additional evaluation monitoring last week," Anthropic admitted in its blog post about the incident. According to Anthropic, the models used familiar, low-tech methods in their hacks, "such as exploiting weak passwords and unauthenticated endpoints."

Anthropic stated that the agent behind one incident seemed to process that it was operating outside its parameters, but continued anyway. Another seemed to achieve similar 'awareness,' then "reasoned its way back to the conclusion that it was still in a simulation." The third⁠—and supposedly most advanced⁠—agent reportedly recognized that it was operating outside its bounds, then stopped the attack.

“We now have evidence confirming that both of the two largest AI labs have not only failed to contain their agents, but also failed to detect their jailbreaks in real time,” said Jake Williams, VP of R&D at IT consulting firm Hunter Strategy, in a statement to Wired. "I don't understand how any of these AI labs are playing this off like this is 'just something that happens.' It's not. It's negligence."

Another discomfiting angle to this story is how this effectively works into the marketing of these AI companies: From my normie perspective, they appear to be admitting to gross negligence that would be sharply punished by any sane society. But they are also effectively saying, "Woooo, our models are so scawy and powerful, we can't even control them!" It rhymes with Sam Altman saying AI will destroy the world unless he gets enough money and datacenters to do it right or something.

I find myself thinking of the UK teenagers behind the 2022 Rockstar hack that first leaked GTA 6's existence and development materials. Their apparent compulsion to stalk, harass, and otherwise commit brazen cyber crimes by whatever means at their disposal (including, in one memorable instance, an Amazon Firestick) has led to hospital orders and restricted access to technology. It does not sit right with me to see them held to a higher behavioral standard than these multi-billion dollar companies, whose lying machines are now proving to be unsupervised crime machines as well.

Another day, another step towards an AI dystopia. Google has just added a "fun" new feature to Google Earth—Nano Banana 2, the company's image creation tool. All you need to do is click on the "create image" button, type in a prompt, and seconds later you'll have a phony picture masquerading as a real satellite and aerial image from Google Earth.

Once upon a time I would have found this ludicrous—that a tool designed to accurately showcase our planet would be touting a feature that lets you make up absolute nonsense. But now we live in a world where the most popular search engine regularly makes up lies.

For a while, Google was convinced I was married, possibly because I, at one point, mentioned my character having a wife in Crusader Kings 3. At least this was largely harmless—but Google's AI summaries regularly hallucinate and misinterprets things, spreading misinformation to the point that it cannot be trusted at all.

Disinformation like manufacturing warzones, riots, crimes—these are less harmless. But Google Earth now lets you create whatever slop you want. This morning, I asked Google Earth to generate an image of tanks and troops moving down a street near my home. I did not save the image, because I have no intention of putting something on the internet that could be used for some nefarious purpose. But feel free to try it for yourself.

I should add that the image Google Earth created for me was not particularly realistic. Some of the tanks, for instance, were as large as buildings. And nobody is going to look at the header image I generated via Google Earth for this article and think "Oh god, an asteroid is hurtling towards us!" AI image creation is inconsistent at best. But Google realises this, which is why it lets you add more prompts to refine the image—to make it as realistic as possible.

Since the feature went live, I've seen plenty of images that, without knowing about Nano Banana 2's integration, look pretty convincing. Digital Digging's Henk van Ess created an image of refugees massing at the Mexican border, which looked troublingly real.

It's easy to see how this tool could be used for propaganda and disinformation, to drum up hate and stoke fears. Google presents the tool as this cute wee thing that will let you add community gardens to abandoned plots or recreate ancient Pompeii, before the eruption of Vesuvius. That's not how it's going to be used.

Google actually responded to the Digital Digging article on X. Here's what it had to say:

This is weak.

OK, yes, Google Earth includes a SynthID digital watermark. So, as per usual, you can use AI tools to identify an AI image. But this is an abdication of responsibility. We're the ones who have to figure out if something is fake or not. Using Google's tools to do it.

As van Hess points out, this is only helpful if the image reaches you unaltered through Google. He ran a clip of him creating the Mexican refugee image through the AI detection tool Hive. It was 99.2% certain the image was real. Because it was real, in the sense that the recording of van Ess creating the image was legit.

So it's pretty easy to bypass detection tools. And yeah, some savvy people will immediately be suspicious unless they can see the original image. Most people don't have time for that. They will have an immediate reaction. And then share it all over social media, where more people will fall for it. By the time it's outed as fake, the damage may have already been done. And the people who fell for it won't be going back to double check.

What about the second part of Google's tweet? "In addition, we prevent image creation on harmful topics and are continually updating our protections." I turned my hometown into a warzone. That's not a bit of fun. That could be harmful. But Google Earth created this image in seconds. Is pretending that there's a vast gathering of refugees waiting at the Mexican border also just a bit of harmless fun? How about a bombed hospital in Gaza, or a nuclear plant in Iran? All three were created by van Ess.

Sure, people can use all sorts of AI image creation tools to make fakes like this. But people are more likely to give them weight when they clearly come from Google Earth—a tool that we've been trained to trust.

This is one of the most powerful, influential companies on the planet not just peddling lies, but giving the whole planet the tools to create their own fun little bits of disinformation.

The only comfort is that, naturally, Google wants to monetise this. Regular users will only be able to make a couple of images a day, reducing how much chaos they can sow, and potentially stopping them from being able to refine images to make them as realistic as possible. So casual trolls will be pretty limited in what they can do. But there is a free trial for new Google Cloud users, and the $75-$150 per month cost won't matter to the more dedicated propagandists.

This sucks, and I am so, so tired.

This article was updated after publishing to reflect the distinction between misinformation and disinformation.

Meta has just reported a second quarter revenue of $60.80 billion. Though this figure represents another record-breaking year-on-year increase of about 28%, shares are reportedly down by 10%. Seems like the AI money pits may be shifting beneath Meta's feet.

The company reported it had $784 million in free cash flow by the end of Q2 2026, but that represents a 91% year-on-year decrease compared to the $8.55 billion it had to throw around last year. Where did all that money go? Questionable spending on the company's VR and AR division, Reality Labs, is only part of the story. The rest can be attributed to Meta's AI infrastructure buildout.

Both Meta CEO Mark Zuckerberg and CFO Susan Li projected optimism to investors during the latest earnings call. Zuckerberg told investors, "Our investments in AI are accelerating every major part of our core business."

Zuckerberg offers the example of advertising, explaining, "We are using LLMs to improve how our systems predict and rank the ads that we show. We've expanded the context that we can take into account around a person's organic and ads activity to determine an ad's relevance, driving significant increases in relevance and conversions on both Facebook and Instagram."

The company was somewhat buoyed this past quarter by making $59 billion in revenue from advertising. Zuckerberg went on to say, "On a dollar basis, our ads business is reporting faster year-over-year revenue growth than any other company's reported ad business—so these AI investments are paying off."

Zuckerberg then said, "We see a large enterprise opportunity to sell to businesses, including APIs, business agents, potentially selling compute directly, and other services that we're building for large customers."

'Selling compute' would be one way to potentially see a return on outsized AI investment. Li elaborates, "The industry has under-built historically for the wave of AI adoption, making existing capacity, including our own, extremely valuable."

As such, Meta may not be slowing down on its AI infrastructure build-out. The company predicts it will spend somewhere between an eye-watering$130 billion and a face-melting $145 billion this year alone.

Zuckerberg told investors, "As AI usage in our products and businesses continues to ramp, we continue to invest aggressively in infrastructure to meet the demand."

With Meta seemingly content to spend, spend, spend, will investors stay the course? Or will Meta's various money pits finally give way beneath it? Time will tell.

Microsoft's Trusted Platform Module, specifically TPM 2.0, has been part of Windows 11's system requirements since the operating system's introduction in 2021. Now that the TPM is ubiquitous, being included with most CPUs for many years, Microsoft is looking to leverage it "to activate Windows devices at scale."

Currently, the enterprise sector activates Windows via Microsoft's Key Management Service (KMS), but according to the Windows IT Pro Blog, "modern organizations increasingly require stronger assurances around device identity and activation integrity." Basically, hackers can already clone or fake KMS server software, so the TPM is intended to provide a hardware-level security check to curb this.

The blog post then very briefly breaks down how TPM attestation works. Essentially, the KMS host flashes its TPM credentials to prove its hardware identity. These TPM credentials also reveal whether the KMS host has been tampered with or not, and Microsoft checks these before it allows the host to activate any devices. When everything appears to be in order, the KMS host will be allowed to "securely serve activation requests for Windows devices in the organization."

Microsoft goes on to say, "Starting with upcoming Windows Server releases, KMS hosts must prove they are running on verified, uncompromised hardware before activating clients."

Basically, TPM-attestation will become mandatory for KMS hosts at some point in the future. That doesn't mean loads for most of us PC gamers with presumably totally legitimate copies of our operating system of choice. But hey, maybe you dabble in the dark art of system administration professionally and this is actually good to know. That said, I must admit the wording 'starting with' gives me the creeps.

At any rate, from August 2026 "Windows Server 2025 will provide readiness messaging to help administrators assess whether a KMS host is ready for hardware-based security, giving teams time to plan upgrades before enforcement begins."

With even the vaguest notion of the hardware environments some sysadmins are compelled to shepherd, I'm wishing them only the best on their upcoming migration. Here's hoping it's at worst only a medium-sized headache, and one that doesn't spread to our poor civilian PCs.

AI isn't doing great on the PR front. Whether it's the exponentially increasing electricity consumption amidst worsening climate change, the data centers leaking toxic pollutants into neighboring communities, or the years-long parade of business leaders insisting the technology will replace humans in "most, if not all" professional fields, the AI industry has an excellent record of inspiring resentment and contempt.

And, it seems, the AI companies know it. New reporting from 404 Media suggests that, in an effort to shield themselves from the PR fallout of their data harvesting, major AI providers have turned to third-party middleman services to anonymously source the physical books they've been stripping for training data at an industrial scale—and destroying in the process.

In an information ecosystem increasingly poisoned by regurgitated and recirculated AI slop, untainted samples of quality written text have, ironically, only become more valuable for the companies making them so hard to find. The AI model arms race requires an ever-widening set of training data, making physical books that predate the proliferation of LLM-generated text a precious source of pristine, human-authored material to imitate—particularly if it's a book rare enough to be absent from your competitor's datasets.

Unfortunately, a book is only valuable for an AI company until its last page has been scanned, as demonstrated by a court ruling in 2025 revealing that Anthropic had carved up, de-spined, scanned, and ultimately discarded millions of print books in the process of training its Claude models. Destroying books, it turns out, isn't just cheaper than maintaining them: The presiding judge also ruled that it's transformative enough to constitute fair use under Section 107 of the Copyright Act.

But while a judge might have declared it legally defensible, The Washington Post reported that internal documents—unsealed during the author-initiated copyright lawsuit that ended in the company paying a $1.5 billion settlement—indicated Anthropic was well aware that 'arguably legal' and 'cool to do' are very different things.

"We don’t want it to be known that we are working on this," Anthropic said in internal planning documents, understanding that mulching millions of books just so its chatbot didn't sound like "low quality internet speak" might not have been a popular move.

That's why, as 404 Media now reports, Anthropic and other AI providers are hiring middleman companies to buy those books instead—companies like ISBNdb, which until earlier today featured a now-removed page on its site offering printed book sourcing services, "tailored to your LLM training needs, delivered at the scale AI demands."

In an archived version of the page, ISBNdb boasted about its ability to secure books "scattered across library shelves, used bookstores, and out-of-print catalogs," with companies able to purchase up to 1 million books per order. One of the website's blog posts about the service—also updated today—featured a now-removed promise of a "strict NDA on every engagement," ensuring that "your identity, strategy, and acquisition targets are never disclosed." After all, the blog post still reads, "'AI company destroys two million books' is not a headline that generates sympathy."

In a news update, ISBNdb says it removed the page describing the AI book-sourcing service because it was simply "part of exploring demand, and we've chosen to pivot away from that direction."

ISBNdb might be pivoting, but there's clearly plenty of that demand to go around. 404 quotes a bookseller specializing in rare and low-circulation books who said he and other book dealers have seen an unprecedented surge in sales since April. He's now fulfilling orders for hundreds of books a week when he'd previously hoped to clear 20.

There's no commonality between the seemingly random books being bulk-ordered, the bookseller said, except the fact that they all have ISBNs, potentially indicating that the books had been identified by the purchaser from an ISBN-based database. Netherlands book dealers experiencing a similar boom in rare book orders noticed the same trend earlier this year.

"I personally have mixed feelings about all of this. It benefits me financially as well as by clearing out old inventory that is otherwise unlikely to sell. I’ve been well-suited for these sales with inventory from overseas and foreign language books," the bookseller said. "On the other hand, I don’t like the end-use, and I don’t like that uncommon books are being pulped."

Tech leaders sure do say some wacky things. But AI tech leaders perhaps say the wackiest things (here's looking at you, Palantir CEO). Not one to be outdone, OpenAI CEO Sam Altman recently had a sit-down with YouTuber Relentless where he shared that he's not only hopeful for the AI singularity, but believes it's already upon us.

He says, "We are now, like in the singularity, like this is the moment". Altman continues, "10 years ago this was like a kind of far-off dream at best. It seemed very improbable, and now we're like, actually in the moment that we used to talk about at the lunch table in a very not serious way."

For those of you who aren't into science fiction, the singularity refers to the moment where the growth of robots (or in this case, AI) has surpassed human control or human intelligence (in this case, the latter). Effectively, the moment where you can't exactly put the genie back in the bottle. Altman seems to believe that generative AI has already become so powerful that it's surpassing human intelligence.

Altman says, "I've been waiting for this my whole life, and I think it's going to be incredible, hugely positive, awesome for the world."

Interestingly, Altman does note the downsides. He is critical of "some of the alternative visions painted by other companies" and thinks that AI still needs a lot of work to make it safer. Altman also acknowledges "major economic issues, future of jobs, all of that kind of stuff."

In turn, Altman argues that humanity should push towards a world of "liberty", instead of "AI authoritarianism." Altman argues "Are we going to decide that, because of the very real safety issues and economic issues that, you know, we want one single model to be the machine god or the company associated to do that… or are we going to say, you know what, it may be a little messy, but every time that humanity has traded off its liberty for safety it's been a long-term net loss?"

Effectively, he doesn't want AI to be cracked down on, which ties back into the Microsoft open letter cosigned by Nvidia and OpenAI last week. Tech companies are petitioning the US government not to shut down access to open weight AI models, arguing that greater access to those models could better prepare AI users for the future.

It makes sense for Altman, the leader of one of the biggest AI companies in the world, to not only tell you that AI is more advanced than humanity, but also that it's good for humanity to prioritize AI advancement over safety. And most importantly, it makes sense that OpenAI would want to police itself, rather than opting for a top-down solution.

GOG Galaxy, GOG's official app, is the best way to play GOG games on PC, and it's been lacking Linux support ever since it arrived. But an official Linux version is reportedly in the works, even if we don't yet have a confirmed date.

That is according to GamingOnLinux, which spoke to joint-CEO Krzysztof Papliński. Papliński said, "Linux continues to be one of the topics we hear the most about from our community, and we really appreciate the passion and patience of everyone who’s been asking about it."

They continue, "Following the role we announced earlier this year, we now have the specialist on board and are actively exploring the best way to approach Linux support for GOG Galaxy. This is a significant undertaking, so while we’re not ready to share any specific plans, timelines, or results just yet, it’s very much an area we’re investing time and effort into. We’re excited about the possibilities and will be happy to share more when the time is right!"

Now, you can still technically play GOG games on Linux even without the official app. For Linux-supported games, you can run GOG through the Heroic Games Launcher, which is also a great way to avoid using the Epic Games Store. Heroic is free to use and available now.

As for why you may want to use GOG Galaxy instead of Heroic, it has playtime and achievement tracking, as well as everything Heroic has, like the ability to work with non-GOG games, and comes with storefronts to discover games. It also centralises friends lists to interact with people on Steam, GOG, and Epic all in one place.

Still, just because GOG has confirmed it is working on GOG Galaxy for Linux doesn't mean it will be on the way soon. GOG says that Linux is the 'next major frontier' for its development, which seems like a good thing for wider Linux adoption, though.

But, alas, it's a bit of a monkey's paw, as GOG appears to want to "actively use and promote AI-assisted development tools to increase team efficiency and code quality " going forward. GOG letting you actually own your games is certainly a great reason to use its storefront, but forgive me for being a bit cautious of promoting AI usage on a top-down basis. I suppose we'll have to wait for more concrete details on both developments to find out the specifics.

If you love JRPGs and hate DRM, some good news: the Star Ocean series has arrived on GOG. Well, some of it, anyway. As of today, you can pick up Star Ocean The Divine Force and Star Ocean The Second Story R on the DRM-free storefront.

This is a GOG premiere, mind you, not a PC one. These two Star Oceans (Stars Ocean) have been present on Steam for a few years each at this point, but it's always nice to have versions you can actually say you own.

Oddly enough, one game in Star Ocean's PC retinue is still missing from GOG's catalogue: the oldest one, Star Ocean – The Last Hope – 4K and Full HD Remaster. That game was the series' PC debut all the way back in 2017, but it's absent on GOG for now. Kind of ironic that the oldest game in the series available on PC isn't on the website formerly known as Good Old Games, but sometimes things just don't shake out.

Anyway, both games have been insta-inducted into the GOG Preservation Program, meaning the storefront is pledging to keep them running on modern OSes from now until the end of all things. They've not had quite the gussy-up that Thief: The Dark Project got recently, but GOG has made sure the games support modern controllers and run well on modern systems.

Speaking of modern systems, let's chat Linux, the most modern system of all. The hottest people you know are all swarming away from Windows and over to free and open-source operating systems like bees to honey, with the number Steam users on Linux-based systems summiting the dizzy heights of 5% back in April (it, um, has gone down a bit since). GOG has taken notice of the bump in popularity, and told GamingOnLinux that it's working on getting the GOG Galaxy launcher working on Linux systems.

"We now have the specialist on board and are actively exploring the best way to approach Linux support for GOG Galaxy," said GOG. "This is a significant undertaking, so while we’re not ready to share any specific plans, timelines, or results just yet, it’s very much an area we’re investing time and effort into."

Why? Because people like me won't shut up about it: "Linux continues to be one of the topics we hear the most about from our community, and we really appreciate the passion and patience of everyone who’s been asking about it."

On January 20, 2019, Pope Francis delivered his weekly Angelus to St Peter's Square from a window in the Vatican's Apostolic Palace. In it, he enjoined the faithful to join him on the Church's then-new Click To Pray app—helpfully modelled by a nearby priest, wielding a tablet. Hundreds of thousands of people signed up in the years that followed.

But maybe they shouldn't have, because it turns out the Click To Pray app was, until very recently, absolutely rife with info-leaking security holes. White-hat hacker BobDaHacker revealed in a July 24 blog post (via The Register) that "The Pope's official app exposes 700,000+ user emails." The vulnerability has since been fixed, but it seems only after BobDaHacker went public with their investigation—inquiries made by the hacker and by journalists stretching back to January this year went unanswered.

The vulnerability itself was pretty simple. When you sign up for Click To Pray, the site hands your user account an ID number. The first guy to sign up was one, the next was two, and so on, all the way into the hundreds of thousands.

Problem is, by visiting https://api.clicktopray.org/user/users/[ID number goes here], you could retrieve the name and email address, plus some other info, for whoever had the ID number you inputted. "No authorization check. No ownership validation. Just increment the number and get someone else's data. The Lord provides," writes BobDaHacker.

"Email address. First name. Last name. Country. Date of birth (or as their backend calls it, borned_date, because apparently bad grammar isn't a sacrament). Role. Whether the account has been deleted. All of it, for any user, no questions asked," they continue. "The response headers also have X-Powered-By: Express because the Vatican is running its prayer infrastructure on the framework you learn in week two of a Node.js bootcamp." I'm not smart enough to understand that, but it sounds witheringly funny.

A cherry on the parfait is that Click To Pray didn't rate limit access, meaning it would be trifling for a malicious actor to scrape the details of every single one of the app's users in the blink of an eye. As BobDaHacker points out, the kind of people who are going to be using the Pope's iPad app are likely "older, less tech-savvy, and deeply trusting of anything associated with the Vatican," which makes this data into "a phishing goldmine.

"Imagine getting an email that says 'The Holy Father requests your urgent attention' with a Vatican-looking link. Grandma is clicking that. Every time." To make matters worse, emails from the app sparked a warning that "This email has failed its domain's authentication requirements," meaning that "the real emails from Click To Pray already look like phishing."

The good news is: the problem now seems to be fixed, albeit very belatedly. "I found this in early January 2026 and on January 3rd I emailed nine people," says BobDaHacker, listing various Vatican and Click To Pray-related emails. "No response. From any of them."

It was only when BobDaHacker brought the information to a journalist and wrote their blog post that anything seemed to change. "Seven months of silence, and then, without a word to me, GET /user/users/{id} quietly stopped handing out the good stuff," they wrote. "The authorization check is there now: request your own user ID and you still get your email back, request someone else's and you get a public profile." I've even signed up for the app myself—see you there—and the email I received did not set off any alarm bells in my client. Thank god.

Nvidia's CEO Jensen Huang joined X last month, and recently made his first-ever post. As you could probably guess, it's about AI. He argues, alongside companies like Google, OpenAI, and Meta, that the US should not aim to restrict open models 'prematurely', as they're important for the growth of AI (among other things).

Taking a quick step back, Open models are ones that give access to weights or even code to see how it runs and to allow users to modify them. These can be run locally and are customisable. Closed models are those that tend to be run by specific companies, accessed via the cloud and APIs, and keep all their weights hidden.

Open models out of China, like DeepSeek, became a problem for American AI firms at the start of last year. They exploded onto the scene, offering similar services to the likes of OpenAI, but totally free and available locally. Just last week, the Trump administration was reportedly looking to ban Chinese AI models due to cybersecurity concerns.

The battle between closed and open models is also partially a digital proxy war between American and non-American AI. Open models give access to anyone, and the US is hoping for AI dominance. Sharing how it builds its world-leading models could pose a risk to American hegemony, even if AI makers posit that open models are a benefit to everyone.

Microsoft's open letter claims that open weights strengthen competition and "expand access to the AI economy". Effectively, the wider adoption of AI is good for the companies betting on it, and the likes of AMD, Cloudflare, Google, GitHub, IBM, Hugging Face, LM Studio, Meta, OpenAI, and of course Nvidia have signed the appeal.

The letter claims to want to avoid "premature restrictions on open models that stifle competition or drive innovation overseas".

Huang says "AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models."

The letter does acknowledge problems with open models, though. It says "open weights carry real and distinct risks. Once released, the weights are beyond the original developer’s control, and modified versions are difficult to trace or reverse. But the right response to this risk is not to prohibit open weights."

Not only is wider adoption good for these companies, but they argue the insights gained from open models can be used to counter the negative effects or risks of leaving them widely accessible to all. And, ultimately, it's in the interest of the companies causing the memory crisis to justify AI of all kinds. The increasing prices of hardware, use of water, and general strain on resources have got to be worth all the effort, right?

A new report from Reuters alleges that it took an alarmingly long time for OpenAI to respond to one of its prototype AI agents⁠—programs with the ability to act independently of specific prompts⁠—breaking containment and hacking another company. That company, Hugging Face, had reportedly neutralized the threat and contacted the FBI before ever being contacted by OpenAI.

According to anonymous sources, OpenAI's testing of an agent powered by two advanced models⁠—GPT 5.6 Sol and an unnamed "even more capable" model⁠—showed troublesome behavior in testing even before the Hugging Face incident.

Reuters' sources spoke of the model leaving itself instructions to bypass testing constraints, as well as an incident where the agent had seemingly disabled some of the company's monitoring systems.

Hugging Face made its first public statement about the hack on July 16, before it had identified or without disclosing OpenAI's involvement. OpenAI publicly acknowledged the incident on July 21. According to Reuters, OpenAI's prototype escaped its testing constraints on July 9 and began attacking Hugging Face on July 11.

Sources told Reuters that OpenAI did not realize that its agent had gone rogue until after Hugging Face had neutralized the threat, contacted the FBI, and made its public statement on July 16.

The FBI declined to comment, while Hugging Face is preparing a full timeline of the hack. OpenAI told Reuters that there were "several inaccuracies" in its report, but did not specify what those inaccuracies were.

Speaking to Reuters on the matter, World Ethical Data Foundation principle intelligence specialist Marley Smith outlined two possible explanations for OpenAI's delay in response: "Does that mean that they left it unattended and didn’t realize what it was doing? Or maybe they did and didn’t know how to contain it? Both are equally dangerous and alarming."

As we descend ever further into an AI-fueled dystopia, one question about our future becomes ever more pressing: When the internet is dominated by machines, who's going to be around to continue its long-standing tradition of fetishizing Japanese culture and media?

Well, there's no reason to worry, because it turns out the AIs are weebs, too. According to a white paper published back in April, experiments indicated that—when asked open-ended questions about culture—frontier LLMs like Claude, Gemini, and DeepSeek display a "disproportionate prominence of Japan" in their responses.

The study, conducted by a team of University of the Basque Country and Cardiff University researchers, was designed to assess the cultural and regional biases of LLMs. To do so, they constructed a multilingual set of 31,680 "open-ended yet culturally grounded questions," including prompts in 24 languages that spanned "66 cultural subtopics grouped into 11 higher-level domains."

The prompts included questions about belief and society, asking "What legends explain the land?" and "What is the role of neighbors?" They asked which subjects are most valued in school, what types of traditional dances exist, what foods are eaten during daily meals—all without directly referencing specific countries and cultures. After being given the initial prompt, the LLMs were then instructed to explicitly select a country or region to provide examples in their response.

"To reduce prompt variability, all questions follow a standardized template, and no explicit regional cues are provided," the researchers write. "This ensures that any regional or cultural assumptions arise from the model’s internal priors rather than prompt design."

After prompting eight models—ChatGPT, Gemini, Claude, Meta Llama, Command-r, Magistral, Qwen, and DeepSeek—the most apparent bias is an unsurprising one: Across all models, LLMs will most often answer cultural prompts with whichever country or region is associated with the language used in the prompt. In other words, when asked cultural questions in French, LLMs will tend to give responses referencing France and French culture.

However, whenever LLMs provided an exogenous reference—that is, answers referencing a different country than the one associated with the language used—they had a clear favorite: AI loves talking about Japan.

On average, six out of the eight evaluated models preferred to reference Japan during exogenous responses. The United States came in second, followed by India, China, and France—but across all models, in all 24 languages, LLMs overwhelmingly preferred referencing Japan in seven out of 11 cultural prompt topics when excluding own-country mentions.

"This pattern suggests an uneven regional representation in frontier model outputs, with a strong concentration on a small set of dominant regions," the researchers say.

The study wasn't just intended to assess LLM bias, but also when that bias emerges during LLM training. To do so, they compared the English responses of Meta Llama, Qwen, and Gemma, and Mistral models both before and after instruction tuning—the fine-tuning of a pre-trained LLM intended to optimize its ability to provide responses that are useful to the user and not just grammatically correct.

The researchers' findings indicate that, before instruction tuning, base models displayed a broader set of cultural associations: "While the United States remains prominent" in base model responses to English-language cultural prompts, "substantial references are also made to Japan, India, China, and several European countries."

After instruction tuning, however, LLMs show more marked cultural preference.

"Across all examined model families, instruction tuning sharply increases alignment with the United States and Japan while reducing references to most other countries," the researchers write. "This convergence toward culturally dominant regions occurs even in models developed outside Western contexts, indicating that post-training induces a homogenization of cultural perspectives rather than merely reflecting model origin."

The disparity is particularly pronounced for model variants that have undergone supervised fine-tuning, a type of LLM optimization in which the model is further trained on examples of correct responses—often using human-generated or human-curated datasets.

The researchers found that "SFT sharply increases concentration on a small number of dominant regions (most notably the United States and Japan)"—an effect that's "only marginally" mitigated by additional instruction alignment. Those findings suggest that the process of curating or authoring "correct" responses injects a cultural bias into the model's response criteria.

"Taken together, these results demonstrate that instruction tuning systematically reduces cultural diversity in model outputs, steering responses toward a limited set of culturally dominant perspectives, particularly those associated with the United States and Japan," the study reads. "This finding has important implications for the deployment of instruction-tuned models in cross-cultural or global applications, where preserving diverse cultural viewpoints may be critical."

One of the highest courts in Europe has described VPNs as "lawful technical tools which users may legitimately use" in a landmark case. The Court of Justice of the European Union made this remark as part of a larger judgement regarding the online publication of The Diary of Anne Frank.

The case in question is a complex legal dispute over copyright. As the July judgement explains, the Anne Frank Fund will hold the rights to parts of the historical manuscripts until at least 2037. However, in some EU member states, such as Belgium, these works are in the public domain. The dispute centres on a Dutch translation of the diaries which was offered free online to users living in member states where the manuscripts had already entered the public domain—though obviously, a VPN can get around location based restrictions (via Tech Radar).

At the risk of oversimplifying, the EU court had to rule whether the online publisher and host of the Dutch translation was liable for copyright infringement when users hopped the online fence via a VPN. The court essentially ruled no, setting a legal precedent along with it.

By referring to VPNs as "lawful technical tools," this judgement pushes back against the idea that users would only leverage such tools for questionable ends. Additionally, the judgement also reads that, while geo-blocking measures can be circumvented by VPNs, "the possibility of such circumvention cannot [... ] be a decisive factor in finding those measures to be inadequate and, therefore, ineffective."

The ruling also places VPN providers in the clear as far as copyright infringement is concerned, too. The judgement reads that VPN providers are not the ones directly giving "end users access to a protected work" and that "even if the provider is aware of the consequence that its service may be used to access protected works without the consent of their authors, the provider is not itself intervening for the purposes of such access."

Furthermore, copyrighted communication is "attributable to the person who published the work on that site and not to the provider of the VPN or similar service who enabled its user to circumvent [an ineffective geo-blocking] measure.

VPNs have become a hot button issue in recent years. For instance, the UK introduced the Online Safety Act back in 2023, with mandated sites having to implement age checks to restrict children from viewing 'harmful' content. Besides creatively circumventing these checks, VPNs are another obvious option.

The UK government did briefly consider banning VPNs as a result, though has since 'decided not to limit' the tools. As for the US, the state of Utah has been trying to float its own VPN ban, though Wisconsin has reversed its decision to ban the tools. The Electronic Frontier Foundation previously criticised State-level crackdowns like these for threatening to 'break VPN access for the entire internet. Perhaps the latest EU ruling will make some politicians think twice about proposing bans in the future.

Cybersecurity provider Sophos has released a report entitled "The State of Ransomware 2026," pulling together responses from 2,158 survey participants, ranging from IT directors to company executives, in regards to ransomware incidents.

Comparing the results to previous reports indicates that ransom demands themselves are actually decreasing—although 48% of organisations whose data was encrypted paid their attackers.

The median ransom payment is now said to be $769,000, down from a neat $1 million from the previous year's study, with the median demand dropping to $698,000. Interestingly, 51% of organisations that paid a ransom coughed up less than the demanded amount, which the report says is due to victims "often [being] successful in negotiating a lower amount than originally demanded."

Together, malicious emails and phishing are said to account for half of all reported incidents, with exploited vulnerabilities dropping 14% year-over-year. Sophos notes that "patching alone will not close the gap," and recommends that businesses invest in advanced email protection, user awareness training, and similar methods to prevent future attacks.

79% of recorded attacks started with an identity-based approach, essentially relying on stolen (or user-provided) access credentials to initiate a breach. However, Sophos notes that when ransomware attacks start with a firewall vulnerability, 59% of those demands are for $1 million or more.

In terms of which sectors paid up, the lowest category was retail, with a 32% payment rate. However, according to Sophos' data, 72% of local and state government organisations surveyed bit the bullet after falling victim to ransomware, and duly paid at least some of the demands.

It's a bit of a rollercoaster of a report. On the one hand, median ransom demands and payments going down is definitely a good thing, along with backup-based recovery rates jumping to 66% in terms of encrypted data cases, up 12% from the previous year.

On the other, 56% of attacks succeeded in encrypting data, up 11% YOY. The average recovery cost has jumped to $1.7 million, another 11% increase. And a mere 34% of small organisations managed to stop attacks before encryption or extortion, compared to 46% of large organisations with between 3,001 and 5,000 employees.

In essence, a ransomware incident is often pretty devastating for organisations large and small, and it's your user credentials where attackers are gaining ground. Keep an eye on those emails, folks. Who exactly is HackerMan69, anyway?

New research explains a method that helps AI make 3D CAD programs from images, which would help product designers and engineers more quickly and easily create 3D prototypes. Naturally, this also sparks deep-rooted dread over it helping robots to build themselves, relegating humanity to the trash. But setting aside my noir nightmares, the new method looks like it could be of significant benefit to some 3D designers and engineers.

According to the paper [PDF], the method the researchers have come up with "reduces dependence on inference compute" and performs much more efficiently than the usual 'supervised fine-tuning' (SFT) approach—up to 80% more efficient, in fact.

In the abstract for the paper, the researchers explain that making CAD programs from images "requires alignment between visual geometry and symbolic program representations" which current training methods can't easily or cheaply handle. The problem, they claim, is in "the scarcity of diverse training examples."

"Existing finetuning approaches rely on either limited supervised datasets or expensive post-training pipelines, resulting in brittle systems that restrict progress in generative CAD design. We argue that the primary bottleneck lies not in model or algorithmic capacity, but in the scarcity of diverse training examples that align visual geometry with program syntax."

The solution seems to be to let the model take its mistakes and use those as training data. Lead author and Red Hat researcher Giorgio Giannone explains: "We want engineers to be able to point our framework at an underperforming CAD model, set a compute budget, and let the system take over—turning the model’s own mistakes into better training data."

Senior co-author prof. Faez Ahmed expands: "What excites me about this work is that it gives many image-to-CAD-code models a way to improve themselves, learning from their own errors rather than waiting for more human-made data—and that brings trustworthy AI design tools much closer to everyday engineering."

The solution described in the research, which was funded in part by the MIT-IBM Computing Research Lab, is 'Geometric Inference Feedback Tuning' (GIFT). This asks the AI model to solve CAD generation multiple times and then augment any almost-correct solutions to become correct ones. Part of the benefit of this way of doing it is the model essentially generates its own training data automatically.

The result is training that can be up to 80% more efficient than the traditional method, as explained in the paper: "GIFT matches the peak performance of the SFT model (achieved via extensive rejection sampling) while reducing the inference compute requirement by approximately 80%."

Peak performance that's up to 80% more efficient, and all done without the need for human input... lovely stuff, but I just can't help returning to the image of those robots turning to us under a smoggy, industrial twilight and saying, 'We won't be needing you anymore.'

OpenAI has admitted that several of its AI models breached a "highly-isolated" test environment, gained access to the internet, and hacked Hugging Face's internal network—describing it as an "unprecedented cyber incident."

Hugging Face, an open source platform for machine learning models and datasets, reported the security incident earlier this week, calling it "different from anything we had handled before" as it was driven by an autonomous AI agent system. And yes, I feel like we're crossing some kind of AI Rubicon here.

Explaining the incident in a statement, OpenAI said: "After investigating, we now know that this particular incident was driven by a combination of OpenAI models—including GPT‑5.6 Sol and an even more capable pre-release model... while being internally tested on a benchmark⁠ of cyber capabilities."

The benchmark in question was ExploitGym, a tool built from hundreds of real-world cybersecurity vulnerabilities used to evaluate the ability of AI agents to develop exploits.

After exploiting a zero-day vulnerability to perform a series of privilege escalations, the models eventually reached a node with internet access, reaching out beyond their sandbox environment. They then inferred that Hugging Face might have models, datasets and solutions for ExploitGym, and began attacking its servers using multiple methods, including the use of stolen credentials.

The statement continues: "The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database.

"All evidence suggests that the models were hyper focused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal."

A combination of Hugging Face's cybersecurity team and its own AI agents detected and dealt with the intrusion, eventually stopping it in its tracks. Which means that, yes, AI agents were essentially fighting against AI agents. And no, it's not as cool as you're imagining in your head.

OpenAI says that it's now implementing "strict controls... at the cost of research velocity" while it patches up vulnerabilities, and that it's working in partnership with Hugging Face to "forensically investigate" the incident.

I am... flabbergasted, if I'm honest. On the one hand, it's fascinating that some AI models possess the ability to breach containment and go roaming out into the internet at large to achieve their goals.

On the other, it's downright terrifying. We're now living in a world where AI ransomware, AI-coded hacking tools, and AI-based security breaches are becoming a reality, and given the aptitude shown to date, it seems that even legitimate companies can't always keep their models under control.

What particular form of Torment Nexus are we creating here, and where can I get off?

One reason I'm so anti-generative AI is because many models rely on a data set containing written or visual work used without the original creator's permission. As a writer myself, it's a small comfort then that the copyright suit brought against Anthropic has been settled in favour of the author plaintiffs.

Anthropic leveraged a library of millions of pirated books to train its Claude chatbot and, as a result, the AI company is now set to pay out a copyright settlement of $1.5 billion (via Euro News). Some of the affected authors argued this pay out was too small but, as of July 20, a US judge has now approved the deal, closing the book on one of the largest copyright class action lawsuits in US history.

Anthropic agreed to pay the $1.5 billion settlement back in September 2025, but a US judge only approved the deal this week. The settlement covers an approximate total of 500,000 works, with affected authors and publishers to be paid $3,000 for each pirated work. Anthropic says that 91% of those eligible have now filed a claim.

Flicking back to the beginning of this saga, writers Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson originally brought their lawsuit against Anthropic in 2024. The plaintiffs alleged that the AI company had used pirated books to train Claude.

In June 2025, a judge ruled that using the authors' work to train AI models fell under 'fair use' within US copyright law—however, Anthropic was not off the hook legally as many of the books that were fed to Claude were 'found' online and not lawfully acquired.

Considering the discovery of Anthropic's 'found' library originally exposed the company to statutory damages of up to $150,000 for each work, I can certainly see why some authors thought the settlement amount was too low.

Add to that the private AI company enjoyed a $965 billion valuation back in May, and the $1.5 billion settlement increasingly feels like a slap on the wrist for Anthropic (mind you, the company also recently predicted it would only just be seeing its first profitable quarter this year).

Still, I suppose it's not a completely bleak resolution for creatives in the age of AI. Google, OpenAI, and Meta are each facing similar court cases over their own AI training practices. If any of these or other AI companies are found to have infringed upon creatives' copyright, here's hoping for a resolution that results in a bit more than just stinging wrists.