Forever Physical: In a world where gaming giants are actively trying to kill optical discs and other physical media, GOG is doing its own thing. The retro gaming store is expanding its dedicated preservation efforts, giving customers a new way to enjoy their growing collection of digital purchases.

GOG recently announced the addition of a few more titles to its Preservation Program, which was introduced to improve support for some classic games sold on the store. The program is now welcoming five new titles, which also come with their own interactive 3D boxes scanned from original packaging.

GOG’s new wave of Preservation Program titles includes Star Trek: 25th Anniversary, Armikrog, Descent, Hitman: Codename 47, and Myst. The five corresponding 3D dioramas were created after scanning the front, back, spine, and all the additional elements of the original boxes. GOG chose the GLB format to share the interactive 3D objects, which should be easy to view with native 3D viewers available on both Windows and macOS.

Furthermore, GOG entered into a partnership with Big Box Collection to offer flat, printable versions of the aforementioned 3D box art. Ideally, users should be able to print, cut, fold, and glue all the art’s elements together to create their own game box replicas. Big Box Collection is a “one-man show” started by a fan of boxed games in 2015. It now offers 3D replicas of around 890 physical games released on PC and other platforms.

As highlighted in GOG’s announcement, game boxes can be affected by the same durability and reliability issues that make optical discs, floppy disks, and other physical media so difficult to preserve. Providing printable scans of the original boxes should help game preservation from a different angle, pleasing people interested in “owning” fully tangible replicas of their favorite titles.

In fact, there is a whole community on the GOG forums that has been striving to add a “boxed” layer to the store’s digital offering since 2008. People have been designing custom “DVD covers” for GOG games for years, and the store is now contributing to this effort with its own “official” selection of printable boxes.

In 2025, I spent hundreds of euros on a brand-new Epson EcoTank ET-8500 printer just to bring my GOG collection to the shelves.

The GOG Preservation Program launched a couple of years ago to ensure that classic games would remain playable on modern systems even after their original developers had abandoned them. GOG is adding its own tweaks, mods, and compatibility improvements to the original releases. GOG has not provided any information about plans to add even more 3D and printable boxes for some of these “preserved” games.

For the first time in more than a decade, someone other than Tim Cook ran an Apple keynote yesterday, and John Ternus saved the biggest reveal for almost an hour in.

We watched it last evening so you didn’t have to – although Apple took its time before unveiling the marquee attraction.

This was the first Apple keynote in more than a decade where Tim Cook wasn’t the one walking us through the big reveals. That job now belongs to John Ternus, and yesterday was his big debut as CEO addressing its major product launch from the company’s headquarters in Cupertino.

Cook did pop up for a cameo, in what felt like a clear passing-of-the-torch moment, but this was Ternus’s show. He set out his stall up front, emphasising that privacy is at the heart of everything Apple does – this is “personal intelligence that’s actually personal”, and private, he insisted. Whether it runs on-device or in the private cloud, there is no ambiguity about where your data goes compared to “other players”, he said. It’s clearly the line Apple wants to be recognised for under this new leadership, and it set the tone for much of the keynote.

iPhone still at Apple’s core

Two new phones anchored the event line-up: the iPhone 18 Pro and iPhone 18 Pro Max. They keep the same design language as last year’s Pro models, though this time round Apple’s added four new finishes to choose from: black, light blue, glacier (a silvery tone) and burgundy.

Under the hood, both phones get the new A20 Pro chip, which is Apple’s first built on a 2-nanometre process. It packs a 6-core CPU, split between two performance cores and four efficiency cores, alongside a 7-core GPU that Apple says delivers 40pc faster graphics, plus a second Neural Engine, all in service of what the company is billing as its most AI-capable iPhone yet.

Battery life gets a bump too; the Pro now matches what the old Pro Max managed last year, while the new Pro Max stretches to a claimed 30 hours. On price, in the US, the 18 Pro starts at $1,199 for a 256GB model, and the Pro Max starts at $1,299, also for 256GB. European pricing is to follow.

‘Digital negative’ for the AI-sceptical era

As the whole world of photojournalism becomes fraught with doubts over the veracity of photos, it was interesting that Apple introduced a ‘Reference Image’ feature in its iOS 27, which Ternus described as akin to a “digital negative” of the original image. In practice, it works by attaching metadata to a photo to certify the image hasn’t been AI-generated or altered. This feature will not initally be available in Europe and China.

On the camera lens itself, photography enthusiasts will love that variable aperture was confirmed on the iPhone 18 Pro line (although not the case on the Duo). Ternus had teased the move yesterday morning with a cryptic post on X.

The new Apple Watch line-up leaned all-in on health this year, with the Series 12 and Ultra 4 both clearly built around that theme. New watch models can use microphones and other sensors to generate an AI recap of your day, and offer a ‘readiness’ reading of what you are ready to take on, in an echo of a prominent feature of another popular wearable, the Oura ring.

The slow reveal – the foldable

It was almost an hour into the presentation before the iPhone Duo even got a mention, long enough that commentators on social media started openly wondering whether it was going to happen at all. But then it dropped almost as an afterthought – the moment everyone had been waiting for, the long-awaited iPhone Duo, Apple’s shiny new foldable phone.

Open it up and, Apple says, you’re holding the thinnest iPhone it has ever made. It has two screens sharing the same aspect ratio, so that whatever you’re looking at scales between them without distortion.

Closed, the outer display measures 5.4 inches and covers 90pc of the screen area you’d get on an iPhone 18 Pro. Open it out and the inner display covers 7.6 inches. Both screens are Super Retina XDR with ProMotion, Always On and a peak outdoor brightness of 3,000 nits. The inner display has a nano-texture finish that cuts down on glare and makes the crease down the middle less noticeable.

The body is built from grade-five titanium, with Ceramic Shield protecting the back and the tougher Ceramic Shield 2 on the front, which Apple rates as three-times more scratch-resistant than the previous generation. It’s rated IP68 for splash, water and dust resistance.

Rather than Face ID, Touch ID is built into the side button, and you can also unlock the Duo with your Apple Watch. The hinge alone is made up of more than 100 components, and there’s no physical SIM tray anywhere – so it’s eSIM-only, wherever you buy it.

It runs on the same A20 Pro chip as the 18 Pro line, on that same 2-nanometre process, with a 6-core CPU that Apple says is up to 20pc faster than the A19 Pro, a 7-core GPU that’s up to 40pc faster, and a new dual 16-core Neural Engine offering double the on-device AI compute of the previous generation.

To keep all that cool, the Duo pairs the chip with a custom vapour chamber, and it runs on a dual-battery set-up with one battery on each side, working together as a single pool. Apple claims up to 31 hours of video playback if you’re mostly using the inner display, up to 44 hours on the outer display, and up to 24 hours if you’re using both about equally.

It also charges quickly, reaching 50pc in around 20 minutes over a cable, or about 30 minutes wirelessly with MagSafe or Qi2.

On the camera, there’s a 48MP main sensor with an integrated optical-quality 2x telephoto built in, so there’s no separate telephoto lens to speak of. It can shoot at the full 48MP or default to 24MP, and it’s paired with a 48MP ultra-wide sensor, as used on the 18 Pro. A new camera sits beneath the inner display to handle FaceTime calls.

There are also a few software tricks that only make sense because of the fold. Smart Take automatically captures a photo the moment everyone in frame is posing, Duo Preview puts a live preview on the outer screen so whoever’s being photographed can check their own framing, and Duo FaceTime lets someone else join a call from the outer display.

On the software side, it runs iOS 27 – specifically, version 27.1 adapted for the Duo, with the interface rearranged to make sense of the fold and a new Split View mode that lets two apps sit side by side on an iPhone for the first time.

Apple playbook

“The foldable launch is straight out of Apple’s playbook – wait for the product to work out its kinks and for the market to offer glimpses of viability, then enter and shape the category,” said Forrester vice-president and principal analyst Dipanjan Chatterjee.

“Remember the Apple Watch? The difference is that foldables still represent a minuscule share of the smartphone market.

“The test of Apple’s ambition will be whether it can expand that niche into a meaningful new segment. Otherwise, the foldable risks becoming a premium-priced ornament in the portfolio, imparting badge value but not meaningful growth. The difference is that a foldable plugs directly into Apple’s extraordinarily well-established iPhone ecosystem, giving it a considerably better shot at commercial success.”

The Duo comes in two colours, ‘star white’ and ‘night sky’, and starts at €2,369. It will be available for pre-order on 16 October in 70 countries and regions including Ireland, the UK, the US, France and Germany.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Multiple social media posts from the US President suggested renaming New Mexico to “New America,” and congressional representatives have preemptively asked Apple and Google not to rename the state.

The United States President has begun using the names on the map as some kind of negotiating tactic in ongoing trade wars he started. After executive orders changed The Gulf of Mexico and Lake Ontario in the GNIS database, Apple Maps reflected that change.

According to a report from The Washington Post, Democratic congressional representatives from New Mexico have asked Apple and Google to not comply if another name change is attempted. The requests have been made after multiple social media posts from the President, but no executive order or official plans have been made.

The request appears to be as performative as the President’s social media posts, because Apple and Google have no say in the matter. The Geographic Names Information System (GNIS) is a database run by the federal government, which is affected by executive orders, and what the companies draw from for local place names.

In using the database, Apple and Google must use the names or risk being barred from access to the system. They can file formal complaints, but that’s about it.

Apple Maps didn’t update Lake Ontario until GNIS was updated and pushed the change to their backend. Google, however, made the change to Google Maps immediately following the executive order being signed.

If the congressional representatives of New Mexico want something done, they could raise the concern in Washington and utilize the powers Congress has at its disposal. Instead, they’ve pointlessly written a letter to companies that have no horse in the race.

The United States President doesn’t have the power to change the name of a state, even with executive order. While he may attempt some loophole, like forcing a map name change without an official name change, no move has been made yet.

In this particular case, these whims may just be more hot air to distract from the numerous issues plaguing the administration. All I would ask of these members of Congress is this: focus on solving the problem instead of going after those that have no control over the problem.

A physical SIM is a small removable card. An eSIM is a chip built into the phone that can’t be removed. The real difference isn’t security. It’s how easily you can switch phones or carriers.

More phones sell as eSIM-only now, and more people travel with a phone that needs a local SIM for a few weeks at a time. Both trends make this comparison worth understanding before you buy a phone or switch carriers.

The short version: an eSIM and a physical SIM do the same basic job. Both connect your phone to a carrier’s network. They just get set up and swapped differently, and the differences matter more in specific situations, like switching carriers to save money, than they do in daily use.

Quick Take

What Each One Actually Is

A physical SIM card is a small removable chip. You insert it into a tray on the side of the phone. It stores your phone number and account details, and carriers mail it to you or hand it over in a store. Move it to a different phone, and your service moves with it.

An eSIM does the same job, but it’s built into the phone itself. It can’t be removed. Instead of inserting a card, your carrier sends the account details to the phone digitally, a process GSMA (the industry body that sets the eSIM standard) calls remote provisioning. That’s also why eSIM technology fits into devices too small for a SIM tray, like some smartwatches.

One thing that doesn’t change either way: your phone’s IMEI number identifies the device itself, separate from whichever SIM or eSIM you’re using. Most current phones support one type or the other, and many support both. Which one your specific phone has, or needs, is worth confirming before you buy or switch.

eSIM vs. Physical SIM at a Glance

Here’s how the two compare on the things that actually affect daily use.

None of these differences make one option flatly better. They just show up in different situations, which is the point of checking the sections below against your own habits. If you swap phones often, the top row matters most. If you travel a lot, the second row does.

How It Works in Practice

Switching to a new phone

With a physical SIM, you take it out of the old phone and put it in the new one. That’s the whole process. With eSIM, Apple’s own instructions list several ways to move a number to a new iPhone, including a direct transfer between phones, scanning a carrier QR code, or activating through the carrier’s app. None of these take long, but they all take more steps than pulling out a card.

Switching carriers

Physical SIM switches usually mean waiting for a card in the mail or visiting a store. eSIM switches are typically faster. Most carriers have an app that installs the new network profile digitally, often finishing in minutes rather than days. Either way, switching only works if your phone itself isn’t locked to your old carrier. If it is, you’ll need to unlock the phone first before a new SIM or eSIM profile will connect.

Running two phone numbers on one phone

This used to require a phone with two physical SIM trays, which most phones don’t have. eSIM changed that. Google’s Pixel phones, for example, support running one physical SIM alongside one eSIM, or two eSIMs at once, with Google specifically pointing to keeping a business number and a personal number on the same device as a real use case.

Traveling internationally

An eSIM lets you add a local carrier’s plan before or right after you land, without hunting for a SIM card shop. A physical SIM means either buying one on arrival or setting up international roaming with your home carrier, which usually costs more.

Common Misconceptions

“eSIM is just more secure”

This is half true. An eSIM can’t be physically pulled out of a stolen phone, and it’s harder to clone than a physical card. But the type of fraud most people actually worry about, someone tricking your carrier into moving your number to a different device, works about the same either way. As NordVPN’s security comparison puts it, SIM-swap fraud isn’t meaningfully less common with eSIM. The security difference is real, but narrower than the marketing suggests.

“eSIM means the phone has no SIM card slot at all”

Not necessarily. Most Android phones that support eSIM still have a physical SIM tray too, so you can choose either or use both. The clear exception is recent US iPhone models, where Apple’s US versions ship eSIM-only with no tray at all, while the same phones sold outside the US keep one. Check your specific phone and region rather than assuming.

“You can’t have a backup SIM with eSIM”

You can. It’s just less convenient. A backup physical SIM can move to any spare phone in seconds. A backup eSIM profile needs the same activation process as the original, which usually needs an internet connection and cooperation from the carrier.

Limitations and Edge Cases

eSIM carrier support keeps growing, but it’s still uneven. Some MVNOs (smaller carriers that lease network access from the big ones) and some countries don’t support eSIM activation yet. This is separate from whether the phone’s network bands match your carrier, which matters regardless of SIM type. Check both with your specific carrier before assuming either will work.

Handing your connectivity to someone else is harder with eSIM. Popping a physical SIM into a family member’s phone during an emergency, or into a cheap backup phone while yours is repaired, is simple. Doing the same with an eSIM means going through carrier activation again, which isn’t always fast.

eSIM activation generally needs an internet connection, whether wifi or another device’s data. That’s a real problem if your phone is lost or dead and you’re trying to activate a replacement with no other way online.

Older phones without eSIM hardware are simply excluded from any carrier or plan that’s gone eSIM-only. If you’re buying a phone specifically to get the lowest cost on an older model, confirm it actually supports eSIM before assuming a factory-unlocked phone purchase will work with an eSIM-only plan.

Key Takeaways

FAQ

from the hi-we’re-from-the-‘how-high’-department dept

There are multiple things this administration is incapable of doing: winning wars, curbing inflation, not acting like Nazis, wearing shoes that fit them, etc. But if there’s one thing it’s most known for, it’s the unwillingness to take a loss and move on.

That’s why the DOJ is still trying to imprison/jettison into a war-torn African nation its first spectacular failure: Kilmar Abrego-Garcia. And that’s why — multiple attempts and several million dollars later — Trump is still pretending what happened to the Lincoln Reflecting Pool were acts of vandalism, rather than the shoddy work of his preferred contractors. Those facts are on the permanent record, supplied by his own DOJ and his handpicked US Attorney (and former Fox News personality, natch) Jeanine Pirro.

The DOJ claimed Doug Burgum’s Interior Department lied to it when it moved to dismiss federal vandalism charges against former Olympian David Hearn. To his credit(?), Burgum has been nothing if not consistent. Burgum continued to spread the lies. And Trump has continued to double-down on his bogus “vandals” claims since the DOJ dropped the charges.

The continuing problems aren’t limited to Burgum’s and Trump’s constant lies. The main problem for David Hearn is that the DOJ can always try to run him through the court system again. While the DOJ made it clear any damage was due to poor workmanship, it refused to dismiss the charges with prejudice — something that would have prevented the DOJ from re-charging Hearn in response to White House pressure and/or a series of hirings/firings meant to replace Pirro with someone more sycophantic.

While the DOJ may have an infinite amount of time and money to blow on stroking off Trump’s revenge fantasies, its lawyers have been learning repeatedly that federal judges have a finite amount of patience for their constant bullshit.

Hearn’s legal team has refused to let this case go simply because the DOJ has refused to let this case go. It has asked the court to bring the DOJ back to explain why it won’t dismiss the charges with prejudice, even after publicly admitting any alleged “vandalism” was simply the end result of shitty pool repair work.

The DOJ still refuses to take this step, which has naturally resulted in DOJ lawyers being berated by yet another exasperated federal judge:

Superior Court Judge Todd Edelman asked assistant U.S. Attorney Michael Spence how he was supposed to look at the president’s statements about David Hearn, one of his attorneys and U.S. Attorney for the District of Columbia Jeanine Pirro as “anything other than [Trump] pressuring your office, your boss and the three of you to charge Mr. Hearn, recharge Mr. Hearn with what you had decided to be a meritless prosecution.” 

“Isn’t this a unique threat of meritless re-prosecution?” Edelman said. 

Advertisement

You’ll notice it’s US Attorney Michael Spence now handling this case, since Pirro has been sidelined (but not officially fired) for refusing to engage in a malicious prosecution on Trump’s behalf. In return for Pirro momentarily demonstrating better judgment and displaying some respect for her office, Trump has repeatedly maligned her in public comments and Truth Social posts.

That leaves Spence with nothing but the dirtiest of dirty work — ensuring the Hearn prosecution remains on life support until Oval Office pressure forces him to move forward with “meritless re-prosecution,” to quote Judge Edelman.

Spence had nowhere to go. Having been painted into the corner by the administration that currently employs him, all he could do was offer up a weak non-rebuttal that kind of throws Trump under the Truth Social bus he’s driving, while being as noncommittal as possible about the future of the Hearn prosecution:

Spence replied only that Trump’s statements “speak for themselves,” while also arguing it was premature to discuss that hypothetical notion because no further grand jury action had happened.

Bro, this discussion isn’t “premature.” Rights are there to be protected, rather than just vindicated after the fact. The court isn’t obligated to allow the DOJ to engage in a malicious re-prosecution before it can draw inferences from the administration’s actions and statements. Courts aren’t just there to clean up the messes the other branches make. They can also block the government from moving ahead with planned rights violations.

US Attorney Spence is saying as little as he can in hopes of saving a job that’s probably not worth having. The future will only hold worse things for anyone who thinks it’s possible to appease both Trump and an apparently never-ending series of annoyed federal judges. 30 seconds of boiler plate editing is all it would take to end the Hearn saga for good. But Spence — and the cowards who work with him — will never close this case because Trump won’t be happy until he’s put someone — anyone! — in jail for botching a job he’d spent months bragging about.

Filed Under: asshats, david hearn, doi, doj, doug burgum, interior department, jeanine pirro, malicious prosecution, refelecting pool, trump administration

Companies: atlantic industrial coatings, greenwater solutions

Anthropic has published its most detailed account of Claude misuse, laying out how people tried to use its models for harm. The company said it detected and shut down these operations. They ran across cyberattacks, state surveillance, influence campaigns, conventional weapons work and biological research.

It set out the cases in a threat intelligence report released on 10 September. The report covers activity its Threat Intelligence team disrupted between December 2025 and August 2026.

The report spans seven harm areas. They are cyber operations, influence operations, surveillance, conventional weapons development, biological misuse, scams and fraud, and illicit distillation. Anthropic said the misuse involved its Claude Haiku, Sonnet and Opus models.

None of the cases involved its Fable or Mythos-class models, it added, with the exception of one distillation case. In each instance the company said it banned the accounts and strengthened its safeguards. It also shared intelligence with authorities and industry partners where appropriate.

Cheaper attacks, not new ones

The largest section covers cyber operations. Its central claim is that AI has narrowed the gap between well-resourced state hackers and lone operators. Anthropic pointed to a hacktivist using stolen API keys, financially motivated individuals, and a state espionage operator. Each ran multi-victim campaigns.

A year earlier, it said, that work would have required teams of skilled people. The company reported breaches completed in two to three hours, with dozens of victims handled in parallel by single operators.

One case, which Anthropic tracks as GTG-20006, involved a Russian-speaking operator running espionage against Ukrainian and European government targets. Diplomatic and defence organisations were among the targets too.

The company attributed it in line with public reporting on the group known as Midnight Blizzard. It said the actor used AI to check whether security products had flagged their malware, then automatically rebuilt it to slip past detection.

Anthropic traced a separate operation to two people it described as undergraduate students in Hunan, China. It said they ran what it called agent swarms against roughly fifty organisations. Those included a Southeast Asian government agency, from which the actor retrieved citizen records.

Surveillance built by a single consultant

Anthropic said state-aligned actors and commercial spyware vendors used Claude to build surveillance systems between January and July. The cases spanned China, Iran and West Africa. It presented one as the most striking. A single subscriber used Claude as the engineering workforce for a platform named Lakana 360.

The company assessed the subscriber to be a Bamako-based consultant working with Mali’s state intelligence service.

The system monitored roughly 25 million SIM cards across all three of the country’s mobile operators. Anthropic said it bypassed a legal requirement for a court order before an operator could disclose certain records. It also ran on local models on-premises, so banning the account did not affect the deployed system.

Other surveillance cases named Iranian actors deploying a malicious Firefox extension that harvested users’ identities from social networks. Anthropic also described a religious affairs intelligence unit in China. It said the unit had shrunk from many teams of analysts to a single office. That office now produces thousands of investigations a month with an AI assistant.

In another Chinese case, the company said Claude scored social media posts by political sensitivity and flagged people for what the operators termed control. The report follows earlier TNW coverage of AI turning up in police search tools.

Jacob Klein, who leads threat intelligence at Anthropic, told Axios that AI was making state surveillance cheaper and more efficient. It was not changing who governments target, he said.

“They’re effectively automating parts of the job within the intel apparatus,” he said. He added that the pattern was no longer theoretical. “Authoritarian states are using AI for surveillance, repression and influence operations today,” he told Axios.

A new category: weapons

The report documents what Anthropic described as a new form of misuse. That is the use of Claude to write software for conventional weapons. It detailed six cases, three in China, two in Russia and one in Yemen.

In the Yemen case, the company said a cell in the north of the country ran three weapons programmes. They included a guided rocket, a multi-stage ballistic missile with a stated range goal above 2,000 kilometres, and a set of missile variants that included a hypersonic glide vehicle.

Anthropic said the actors used Claude Code in place of software engineers to develop guidance and control code. It added that they test-fired a guided rocket, a test that appears to have failed.

The Russian cases included freelance actors working on an autonomous kamikaze drone swarm.

Biology, and the judgement calls

Anthropic said it presented five cases of people using its models in ways that could support biological weapons development. It stressed how hard those judgements are to make.

The company withheld the names of the institutions, the countries and the specific biological agents involved. It said the individuals were working scientists, and it did not assert that they intended harm.

In one example from May, it said a request for help writing a grant application involved gain-of-function research on the chikungunya virus. That work was intended to be carried out at a military research institute.

“You are not seeing someone in a comic book kind of way say, ‘Hey, I want to build a biological weapon to kill everybody,’” Klein told The New York Times. “It’s an incredibly nuanced situation.”

On the chikungunya case, he told the paper the company did not know whether the research was meant to be weaponised, but that a military institution doing gain-of-function research was concerning. Anthropic said older models such as Claude Opus 4 were well below the level where they could meaningfully assist such work.

It said it had launched more recent models with stronger safeguards. That shift echoes an earlier report in which the company ran its bioweapon classifiers off during testing. The distillation thread, meanwhile, runs back to White House claims that Moonshot distilled Anthropic’s Fable.

The Chinese distillation cases

The report also expands on illicit distillation. Anthropic defines that as covertly extracting a model’s capabilities to train a rival.

The company said it had identified campaigns from seven China-based labs, all targeting its generally available models rather than Mythos. It named Moonshot AI, which it said silently forwarded customer requests to Claude and displayed the responses as though they came from its own Kimi model.

Over one ten-day period, Anthropic said, Moonshot relayed almost 300,000 requests through a network of 5,380 fraudulent accounts. It attributed more than 23 million exchanges to Moonshot between May and July. The company made similar allegations against DeepSeek, Zhipu, Xiaomi, SenseTime and MiniMax.

Some of the relayed queries exposed sensitive user data, it said, in one instance live credentials tied to a Russian defence agency.

The distillation findings land alongside a US intelligence advisory naming six Chinese firms and Beijing’s rejection of those claims. Anthropic said most of the influence operations it caught drew little or no authentic engagement before it disrupted them. Other platforms have made the same caveat.

Anthropic said it published the report out of an obligation to disclose the misuse, and to give governments and civil society a clearer view of how such threats take shape.

OpenAI has asked members of Congress in recent weeks for clear guidance about whether orchestrating an industry-wide slowdown on frontier AI development would be legal, people close to the company tell WIRED.

Substantive coordination on safety between AI labs may risk running afoul of antitrust law, the people say, which poses a significant obstacle to bringing major tech giants on board with the effort.

Last weekend, OpenAI’s chief scientist, Jakub Pachocki, published a blog post arguing that the best path forward for the AI research world includes “coordinating to slow down future development,” which he thinks will be key to ensuring that self-improving AI systems are safe. In the short term, he expects “voluntary slowdowns to become commonplace until shared safety bars are established.”

But some legal scholars say such an effort could conflict with US antitrust laws. Nicholas Felstead, assistant director of the Australian Competition and Consumer Commission and a former AI policy fellow at the Center for Law & AI Risk, argued in a March article that a coordinated pause in AI development may amount to companies restricting output, potentially violating the Sherman Antitrust Act. It would depend “entirely on the precise details of any agreement,” Felstead wrote. “But even if most safety collaborations would ultimately survive antitrust scrutiny, legal uncertainty can act as a powerful deterrent.”

OpenAI did not respond to a request for comment ahead of publication.

There are some early signs that Congress is listening. In July, a bipartisan, bicameral group of lawmakers introduced a bill titled “Collaboration on Adversarial Threats and Security Risks Act,” which would explicitly permit AI labs to coordinate on security and safety work without the risk of violating antitrust statutes. The House version was referred to the Judiciary Committee but has yet to be taken up.

Caleb Knapp, director of government affairs at the nonprofit AI Policy Network, which endorsed the bill, says it would create legal channels for AI labs to work together to address safety and security incidents. Knapp adds that Congress has a “growing appetite to get something done” on AI safety but says enacting anything into law may have to wait until after the upcoming midterm elections.

While some AI executives may have genuine concerns about antitrust, another camp of AI leaders argue those worries are just convenient cover to avoid discussing the real reasons that AI developers may be wary about collaborating, which go far beyond potential legal liabilities.

For one, AI is a massive business, and these companies are fiercely competing to capture a slice of the nascent market for frontier models. Some executives share the Trump administration’s view that staying ahead of China in AI is crucial for national security. Perhaps most importantly, different AI developers have vastly different opinions about the best way to develop safe artificial intelligence, and therefore may be reluctant to work with one another on the issue.

“First step is for industry leaders OpenAI and Anthropic to stop feuding and work on a pacing proposal together,” John Schulman, an OpenAI cofounder who is now chief scientist at the rival AI lab Thinking Machines, wrote in a post on X earlier this week. “They’ll cite antitrust, but that’s fake—antitrust prohibits certain agreements, but not from jointly developing a proposal.”

Long-simmering fears about the race among AI companies to build and release ever more powerful models erupted into the national spotlight this summer. This week, former Anthropic and OpenAI researcher Jacob Coxon added to the growing alarm when he issued a stark public warning that AI developers were putting humanity at risk. In recent months, a number of security incidents, including OpenAI’s agents hacking Hugging Face, highlighted how the industry’s safeguards have failed to keep pace with the rising capabilities of models. The succession of events has led many lawmakers to make urgent calls for AI regulation.

This is an edition of Maxwell Zeff’s Model Behavior newsletter. Read previous newsletters here.