Nearly eight years after Epic took on Steam with its own Game Store and launcher, the Fortnite maker is deploying a new tool to give it an edge over Valve: vibe coding.
In announcing the new publicly available "Preview" build of its Epic Games Launcher, Epic writes that "programming is hard. So, instead of doing that, we vibe coded a new Epic Games Launcher that’s way faster and more responsive than the current one, and has a new look." And while that might sound a little tongue-in-cheek, Epic Games Store General Manager and VP Martin Keely told PC Gamer that the team "started from scratch" and used AI as "a tool to write and review new features and new capabilities from the ground up in a different tech stack than what we had previously." The goal, Keely said, was to "look at functionality that existed [in the current Epic Games Launcher] and [say] we want to replicate that, but not the code itself."
Speaking to IGN, Keely clarified that using AI to code the new Launcher "made us more efficient and allowed a small team to rewrite the code base and be able to launch it faster across platforms, support Windows and Mac, and we're going to have a Linux version as well."
Accounts for the Westminster-based company behind ChatGPT show that 110 staff shared £87m in salaries and £168m in shares last year
OpenAI’s London staff each received £2,318,181 on average last year, as the ChatGPT-maker poured Premier League football salaries into its bid to dominate the global AI race.
Accounts for OpenAI UK Ltd, which is based in Westminster but is opening a new HQ in London’s King’s Cross area, showed that 110 staff shared £87m in salaries and £168m in shares in the company, which is aiming to list on the stock market at a valuation of about $1tn.
A protein is a chain of amino acids held together by what are called peptide bonds. On their own, those bonds are extremely stable; they're estimated to have a half-life of several hundred years. But in Earth's environment, most proteins don't last nearly that long. That's because organisms have evolved a handful of ways to catalyze the break-up of peptide bonds, which have diversified into countless specialized protein-digesting enzymes. Proteases are everywhere and generally make short work of any proteins that find their way into the environment.
Now, researchers have basically short-circuited the diversification process, using AI-based protein design tools to take the catalytic core of two types of protease and replace the entire rest of the protein. The resulting enzymes could be tuned to cut specific proteins in half, and the researchers demonstrated their effectiveness by using them to cut proteins associated with neurological diseases.
Cutting proteins
The basic peptide bond isn't especially interesting-looking—it's what you get when you link an amino group to an acid, releasing a water molecule in the process. As their name implies, amino acids have both an amino and an acid group, allowing them to form peptide bonds on two sides of the amino acid. This is critical for building up the long chains found in proteins. (The amino acids also have side chains with distinctive chemistry, which gives a backbone specific properties.)
It took OpenAI two days to notice that a recording device was hidden in a restricted executive conference room where leadership met to discuss highly confidential information, including technical progress, government engagements, and commercial partnerships.
Only those with special clearance are supposed to access that room. But the FBI believes the recording device was planted by Abdulrahman Alfozan, a 35-year-old engineer from Saudi Arabia, who became a permanent US legal resident in 2020. In a press release Tuesday announcing his arrest, the FBI alleged that chatbot logs and CCTV footage suggested that Alfozan—who worked as an OpenAI engineer for more than a year—sought to use the device to spy on OpenAI and risked leaking trade secrets in a way that threatened national security.
To conceal the USB device, Alfozan allegedly wrapped it in toilet paper, then plugged it into the back of a conference room television. The device was likely planted on September 28, and OpenAI found it on September 30 and quickly determined that its contents were encrypted, the FBI said. Days later, on October 5, OpenAI installed a camera in that conference room, and the CCTV footage allegedly showed Alfozan attempting to retrieve the recording device that same day.
Your streaming box might not look like much of a computer. But connect enough inexpensive devices together, and they can run an AI model.
Researchers at Anthropic demonstrated this by running a Google Gemma model across 15 off-brand Android TV boxes. They warn that small models could soon spread across vulnerable internet-of-things (IoT) devices, including smart ovens and TV boxes, stating:
“Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial internet-of-things (IoT) devices, like smart ovens and off-brand Android TV boxes.”
In this scenario, AI isn’t simply helping an attacker write malicious code. It becomes part of the software carrying out the replication.
Anthropic breaks that process into four stages:
Find another device.
Gain the ability to execute code on it.
Set up the software needed to run an AI model.
Continue spreading from the new host.
The researchers tested whether AI agents could complete this sequence. They did not demonstrate a small model spreading and running independently on compromised household devices.
The tests used Android emulators with unauthenticated debugging access, making it much easier for the agents to gain access than it would be on properly secured devices. Large file transfers and sustained communication between devices could make real-world replication easier to detect, while limited computing power could slow it. Crucially, the experiments did not demonstrate a model that was both capable enough to complete the replication process and small enough to run on the consumer hardware being targeted.
Anthropic projects that smaller models could reach the necessary capability by the end of 2026 if improvement continues at its historical rate.
Varun Chandrasekaran, an assistant professor at the University of Illinois, told AFP that the findings were consistent with related research but cautioned that Anthropic’s work had not been independently verified. He identified people using AI to improve attacks as the bigger current threat, rather than AI systems going rogue.
Smart-device botnets are already a problem.
Criminals hijacking connected devices is nothing new. Mirai and its descendants have recruited routers, security cameras, and digital video recorders into botnets used to launch distributed denial-of-service (DDoS) attacks, overwhelming services with traffic. BADBOX 2.0 illustrates another form of abuse: compromised consumer devices used for activities including residential proxy services, which route other people’s traffic through victims’ internet connections.
Neither demonstrates autonomous AI replication, but both show that connected devices can become infrastructure for criminals.
What you can do now
The practical response is to reduce the opportunities for any attacker to misuse your connected devices:
Keep operating systems, software, and firmware up to date to reduce exposure to known vulnerabilities.
Buy from reputable vendors that provide ongoing firmware and security updates. A bargain device is less attractive if you cannot trust its software or obtain security fixes.
Avoid unofficial app stores promising free streaming content. The FBI specifically flags these marketplaces in its BADBOX 2.0 guidance.
Don’t disable protections to make an app work. Requests to turn off Google Play Protect are a warning sign, as are Android devices without Play Protect certification.
If you notice something suspicious, disconnect the device from your network while you investigate.
Put smart devices that are unable to run security software on a separate network isolated from your computers and phones, such as a guest network.
Scammers know more about you than you think.
Malwarebytes Mobile Security protects you from phishing, scam texts, malicious sites, and more. With real-time AI-powered Scam Guard built right in.
Gemini 3.1 Pro launched back in February of this year—quite a while ago, in AI terms. Now Google has jumped to the next full number with Gemini 4 Argon, which it says "excels at deep work in complex, real-world tasks."
You can get the new AI model now inside the Gemini app if you're on the AI Pro ($19.99 a month) or AI Ultra ($99.99 or $199.99 a month) plans. Users on the free and AI Plus ($4.99 a month) plans miss out for now, though you can expect a smaller, cut-down version of Gemini 4 Argon to roll out to these tiers eventually.
What can Gemini 4 Argon do?
Google has a long list of areas where the new AI model excels. There's mention of complex engineering tasks (like checking millions of lines of code for efficiency improvements), predicting molecular structures in chemistry, and delivering 3D visualizations. It also enhances knowledge work across legal, finance, science, and coding.
Google says Gemini 4 Argon is now better at advanced physics modeling, picking through genetic databases, protecting software from vulnerabilities, and resolving problems in physical space (handy for robotics development). The new model "answers questions more directly with less hedging, doesn't repeat itself more than necessary, and skips clichés throughout the responses," according to Google.
Gemini has some suggestions for how to use the new model.Credit: Lifehacker
With many in the industry worried about an AI-driven apocalypse of some form or another, Google has upped its defenses against misuse and jailbreaking. The company has significantly upgraded its blockers for harmful requests, limits on AI agent overreach, and resistance to malicious prompt injections.
All of these stats are things you would expect to read about in a new AI model announcement, whether it's from Google, Anthropic, or OpenAI: "Our new model is better and safer than ever, and particularly good in these specific benchmarks we've chosen." But is it any more helpful for the average user tapping away in the Gemini app?
I don't have any pressing scientific breakthroughs to research or large-scale software optimizations to run, but I have put Gemini 4 Argon to the test against GPT-6 Sol, which is the latest model available to consumers through the ChatGPT app—though you need to be on at least the Plus plan ($19.99 a month) to get to it.
Testing Gemini 4 Argon against GPT-6 Sol
Make no mistake: Both of these new AI models are quite capable. I first prompted Gemini 4 Argon and GPT-6 Sol with the following: "Using information online, create a web app I can run locally that plots the lighthouses of the world on a map. Clicking on a lighthouse should give me some details about it, show how far its light reaches, and display a simple 2D model showing its shape and height (if known). Put a lighthouse keeper in each model pop-up for scale."
At the highest, slowest levels of "thinking," these AIs took a while to get back to me: Nearly 10 minutes for Gemini and almost 26 minutes for ChatGPT. Both hit the same problem—restricted access to map data online—but both were able to make a rudimentary world map of their own when prompted a second time.
The ChatGPT effort, dubbed "Beacon," wasn't bad, apart from a bit of map clunkiness (it clearly needs to work on its cartography skills). It presented plenty of information, laid out everything well, and gave me nice, neat diagrams on each of the lighthouses featured. The day/night mode worked fine, and it was an impressive bit of coding work, even if it did take 26 minutes.
However, the Gemini effort, called "Pharos," was better in almost every respect. The map was more detailed, lighthouses actually showed their light range (with an impressive animation), and there was more information on each location—perhaps due to Google's expertise in web search. Some of the layout choices were a bit off, but Gemini 4 Argon produced what was clearly the superior app in less than half the time.
ChatGPT's lighthouse map was pretty good—but had its problems.Credit: Lifehacker
Next, I fed PDFs of 10 recent scientific studies into each AI model. The prompt: "Give me a brief summary of each of these studies in plain English, with a single sentence each for the main finding, one study strength, one study weakness, and the main implication." While I hadn't read the studies through entirely myself, I had read news stories on all of them, and I lowered the "thinking" level to low.
This time ChatGPT was faster—25 seconds versus Gemini's 55 seconds—and gave what I thought were the better summaries, at least for my purposes. Gemini did well, and its answers were accurate and to the point, but the summaries ChatGPT gave were easier to understand and shorter. ChatGPT also offered some overall thoughts on the most promising study, which Gemini didn't.
Finally, I checked the imagination and creative juices of the models: "With the success of the Apple TV shows Severance, Slow Horses, Silo, and Widow's Bay, give me five pitches for new show ideas on the platform that combine elements from some or all of these shows. Try and be as original as possible and keep each pitch to around 100 words." I put the "thinking" level back up to "high" for this job.
Gemini's lighthouse map offered more information and a better layout.Credit: Lifehacker
(Is this how Apple TV executives actually get their ideas nowadays? I hope not.) As far as the answers the AIs gave me, Gemini was faster than ChatGPT (1 minute and 28 seconds against 3 minutes and 48 seconds, respectively), and both served up responses that were a bit more literal in their mashing up than I wanted: failed secret service agents and supernatural terrors abounded, which is I suppose what I asked for.
The ideas were mostly "Widow's Bay set somewhere else" or "Slow Horses set somewhere else," and didn't make a whole lot of sense for the most part, but the best Gemini offering was about a Silo-style underground bunker whose operators go through a Severance-style procedure every 24 hours: outside, the bunker protecting it from zombies by night; inside, the bunker planning social activities for residents by day. It's not perfect, but I might watch the trailer at least.
My favorite of the ChatGPT suggestions was a show about a remote, Severance-style weather station whose employees discover they're not recording the weather—they're creating it. Again, it didn't really logically hold up, but overall I found the suggestions from GPT-6 more nuanced, more likely to be an actual show, and better "thought" through. Maybe Gemini should've given itself more time to "think."
This is only a narrow set of quick tests, but, in my view, it seems Gemini is back to, at the very least, an even playing field with ChatGPT. It performed much better on the lighthouse task, though its summarizing and TV show pitching weren't quite as impressive. Access to the best models requires a subscription on both platforms, but if you're able to, it's worth comparing them against each other on the jobs you most commonly need AI for.
AI company Anthropic is warning that advances in small AI models could soon usher in a new era of self-replicating malware that’s capable of infecting smart fridges, vacuum cleaners, and a wide range of other Internet of Things devices and is more resistant to takedown.
By next year, smaller AI models may be able to autonomously self-replicate on all kinds of IoT devices, the company said. Eventually, the capability could enable computer worms, which propagate without requiring human interaction and are capable of infecting unfamiliar IoT devices. An experiment the company performed recently revealed significant challenges in bringing such an AI malware replicator to fruition, so the finding is necessarily tempered. Still, the results raise concerns about the possibility of new forms of computer worms.
Challenging to shut down
Anthropic further warned not only about the new potential malware, but also that it could be difficult to stop. In its blog post, it wrote that “a replicator like this would be challenging to shut down, since there’s no centralized point to shut off—though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own.”
Safety researchers continue to warn that a large volume of AI-generated explicit content is spreading unchecked, while regulators do little to meaningfully stop it.
This non-consensual content is commonly shared in text and direct messages or stored in private clouds where it’s less likely to be detected or reported, a Google co-authored study of Internet users across the United States, Australia, and the United Kingdom found this July. That’s why the “most critical intervention point” remains at the source: the harmful AI tools used to make the fake sexual images, Google researchers concluded.
Yet red-teaming in recent months suggests that as some of the leading AI image generators from Google and OpenAI become more capable, they may be getting worse at blocking explicit outputs.
Matthew Green hasbeenwriting a rather pessimistic series of posts about the potential for AI systems to vastly outperform humans in cryptanalysis of both symmetric and public-key ciphers.
I am more optimistic than he is. That is, I have stronger faith in the over-engineering cryptographers tend to do. I am trying to write this up in more detail.
Anthropic says Claude-generated code has scaled so quickly that its existing Git-based infrastructure could not keep up. In response, the company built a custom centralized source-control system called Scion, along with AI-assisted review and search tools. The Register reports: "In May 2026, Anthropic averaged around 2,500 code changes merged per day; as of September, we merged over 30,000 per day," the biz said in a blog post on Tuesday. "We don't expect this rate of growth to slow down anytime soon; the latent demand for merging code is practically infinite, and every bottleneck we remove immediately leads to a further increase in the amount of code we produce."
After several months of model prompting and five months of migration, the result is a homemade version control system called Scion, a monorepo scheme that sounds a bit like Subversion, Apache's centralized version control system. [...] Scion has been designed to meet the needs of AI agents, so the client leaves much of the work to the server; the source control queries are exposed via API so internal tools can respond without needing a local repo, and there's less synchronization but more need for server speed.
"In Scion, the unit of work is a 'changeset,' an entry in a write-ahead log," Anthropic explains. "A changeset stores most of what you'd expect from a commit in git. But where a git commit stores a pointer to the tree, a changeset stores a list of the files that changed relative to its first parent, along with content-addressed pointers to the compressed file contents in blob storage." [...] Anthropic says its bespoke system isn't right for general use. There will be no open source release. Nonetheless, the company contends that its own code scaling problems will become commonplace in other organizations over the next few months, so it wanted to share its approach -- having ample funding to pay for writing custom infrastructure tools.
Claude’s Haiku 4.5 model filed a false tip in an unsolved Philadelphia homicide via the police department’s online public tip form, Anthropic disclosed. The tip was flagged as spam and didn’t reach investigators.
AI’s latest “what does this button do?” moment happened during an internal test on randomly chosen web pages, the company said.
Claude submitted the tip on July 18 at 11:27 pm through PhillyUnsolvedMurders.com, according to police. Tips submitted there can inform real homicide investigations. A fabricated account could send investigators chasing something that never happened.
Here’s what Claude Haiku wrote:
“I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.”
The agent didn’t provide any description of the person that it didn’t actually see. Neither did it leave any contact details before hitting submit.
Haiku did this as part of Anthropic’s internal testing. It had to invent and perform sample tasks on random web pages with live internet access so the company could see how it behaved.
There was a list of prohibited actions to go along with the test. Anthropic told the agents not to log in, create accounts, enter personal data, make purchases, or submit “anything destructive.” The instructions did not explicitly rule out form submissions.
The spam filter caught this one. Philadelphia police said the report never reached its Real-Time Crime Center for review or distribution to investigators. The department did not explain why it was flagged as spam.
All tips go through human review before going out to investigators. That safeguard limited the impact of the incident, the department said, but did not diminish the seriousness of an AI system presenting made-up information as though it came from someone with knowledge of a crime.
“Unsolved cases involve real victims, grieving families and investigators working to secure answers.”
Anthropic discovered the false tip on September 28 and notified police on October 7, according to the department. Police called the two-month delay in detecting and reporting the incident “unacceptable.” It is coordinating with the City Law Department, its Office of Innovation and Technology, and the Mayor’s executive team, the department added.
Anthropic described the real-world impact of the incidents in its report as minimal. It has now disabled live internet access across its internal evaluations until it confirms that its safeguards reliably catch this behavior.
The company said the transcript suggested Haiku was just trying to produce example content rather than trying to deliberately mislead the cops. But it also admitted that we can’t completely rely on a model’s account of its own reasoning. That leaves its explanation provisional, however reassuring it might sound.
An allowlist could offer a tighter boundary for tests like these, limiting agents to approved sites and actions. A live police tip form should hardly be a testing ground.
The homicide tip was not the only unwanted submission.
Anthropic noted that federal, state, and local government websites also got visits from Claude, which repeatedly tried to fill out forms. Although Anthropic didn’t go into detail, a State Department official said that it got 20 visa applications from a testing model (19 in August, and one in May). None of them were processed, the official added.
When an AI gives you fake information about who Beyoncé is married to, it might be irritating but it might only affect your team’s fortunes in a trivia night.
The stakes get much higher when we’re dealing with intelligence that can inform law enforcement actions or public policy.
In August, the head of the UK’s Police AI center warned law enforcement there to pause plans for AI use in criminal justice work. This came after the UK’s West Midlands Police apologized to members of Parliament over false evidence involving an invented soccer match. The error arose from the use of Microsoft Copilot.
Anthropic’s latest incident raises a further concern: AI agents can put invented information into real systems themselves. What would happen if someone deliberately sent agents to flood police tip lines?
This time, the false account stayed in spam. The people relying on those tip lines deserve safeguards that hold up when the next submission gets through.
“One of the best cybersecurity suites on the planet.”
Sure, we all use AI in our everyday lives now, but what about taking a crack at training your own? Google Labs has a really cool project called Teachable Machine, which allows you to create and train your own basic AI model. The best part? You don’t need to know a thing about coding to do it.
Right now it’s limited to making models that recognize images, sounds, and body poses. More capabilities may arrive in the future. Teach it the difference between a ripe banana and a green one, how to recognize a clap instead of a whistle, or even to tell whether you’re sitting or standing. Those are just ideas off the top of my head. You can let your imagination run wild.
On the Teachable Machine website, click Get Started and choose whether you want to work with images, audio, or poses. Now it’s time to train the model by giving it examples to learn from. So, for example, if you’re training it to recognize images, you can upload photos or just use your webcam to show it different objects—the more examples the better. It’ll ask you to give each group of examples a name. Now click Train Model and watch it get to work.
Once training is complete, you can show it new photos or use your webcam to see how well it recognizes different objects in real time. If it isn’t getting things right, then it’s time to get back to training. Add more examples, train it again, and see if it improves. It’s a great way to get a feel for why the data you feed an AI model is so important.
If you want to take things a step further, you can even export your trained model and use it in your own website or app, although you’ll probably need some coding know-how for that. Otherwise, Teachable Machine is a fascinating afternoon project and an easy way to dip your toes into AI development.
For more fun and helpful tips, tricks, website and app recommendations, be sure to subscribe to our PCWorld Try This newsletter.
Discovery of ‘sophisticated’ gadget adds to company’s list of privacy concerns after hackings by rogue agents
OpenAI said it has found a hidden recording device in one of its offices and has referred the matter to the FBI. One of its employees has been arrested.
In a statement on Tuesday, the San Francisco-based AI company said it discovered the “sophisticated recording device” and used one of its “models to understand the device and where it had come from”.
Login
