Hayden Field | The Verge

AI leaders sign a statement asking the government to do something about automated AI

2026-07-28T19:14:53-04:00

Employees of OpenAI and Anthropic, as well as Google, Meta, Thinking Machines, Microsoft, Mistral, and other leading AI labs, have written a statement to the US government supporting a potential slowdown of sorts for frontier AI development — or at least a speed-up of global coordinated governance efforts.

“Al could help create a dramatically better future, but that outcome is not guaranteed,” the employees wrote in the statement. “The world’s leading Al companies believe they could be close to automating Al research. It is hard to predict exactly how much this will accelerate Al progress, but there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems.”

The news follows a high-profile cybersecurity incident that rocked the AI industry last week, in which an unreleased OpenAI model escaped its internal sandbox, finagled its way into gaining internet access, and then hacked competing AI lab Hugging Face.

In the public statement, the employees went on to underscore the importance of some sort of coordinated regulation worldwide, so that the ever-intensifying AI race doesn’t outpace safety and monitoring tools.

“To realize Al’s potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight,” they wrote. “But each company—and country—is under intense competitive pressure not to unilaterally slow that acceleration. And today, the world lacks the technical and governance tools to deliberately pace frontier-wide progress. Building on work already underway to monitor frontier model releases: We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated Al development.”

Some of the OpenAI executives or cofounders who have signed the statement include chief research officer Mark Chen, chief scientist Jakub Pachocki, cofounder John Schulman, and cofounder Wojciech Zaremba. As for Anthropic, cofounder Jack Clark, cofounder Chris Olah, cofounder Ben Mann, and chief science officer Jared Kaplan have all signed, as well as Boris Cherny, who created and leads Claude Code, and Ethan Perez, who is Anthropic’s alignment team lead. OpenAI’s former chief futurist, Josh Achiam, and the former colead of its now-defunct superalignment team, Jan Leike, both also signed.

You can see the full statement and a list of the more than 1,100 signatories on a dedicated website titled Pacing the Frontier.

Anthropic releases Opus 5 with ‘close’ to Fable 5’s capabilities

2026-07-23T14:33:59-04:00

Weeks after Anthropic’s latest toe-to-toe with the US government, and days after an OpenAI security incident that dominated tech industry discussions, Anthropic on Thursday released its newest model, Claude Opus 5.

The company said in a release that Opus 5 “comes close to the capabilities of Claude Fable 5 in many domains” and is much better at complex coding tasks. (Fable 5 is the public-facing Mythos-class model that drew the government’s ire, was taken offline for a few weeks along with Mythos 5, and then brought back with even stronger cyber safeguards than before.)

The Fable 5 concerns — and the ensuing weeks of negotiations between Anthropic and the government — kicked off a new era of AI regulation for the Trump administration. Soon after, when OpenAI released GPT-5.6, it underwent a roughly two-week trial period of sorts, rolling out in limited preview solely to government-approved entities.

Anthropic promises stronger cybersecurity safeguards with Opus 5

When asked whether Anthropic ran Opus 5 by the Trump administration before this public rollout, Anthropic spokesperson Danielle Ghiglieri told The Verge, “We continue to work with our government partners to conduct their own independent testing of our models. This includes Opus 5.”

Anthropic is marketing Opus 5 especially to its enterprise customers, saying it’s best for knowledge work and biology and that it’s the ideal everyday model, with Fable 5 still being best for the most ambitious types of work and long-horizon AI agent projects. But the company was careful to say in a release that it has stronger cybersecurity safeguards than its predecessor, Opus 4.8, and that it’s “the most aligned Opus model and is the least susceptible to being tricked into misuse.”

With the new model’s release, the company also seems to subtly address recent controversies over allegedly hidden usage caps — which have led to at least one lawsuit — and the AI money squeeze, or concerns that AI labs’ costs are being passed onto their customers in new ways. Per million tokens, pricing for Opus 5 is the same as its predecessor ($5 input / $25 output), and Anthropic made the point that it’s “half the price” of Fable 5. It’s also slightly cheaper than OpenAI’s GPT-5.6.

Customers can also choose “Fast mode” for Opus 5 (a feature that’s being launched today in research preview), to get higher speeds for double the price. There’s also an option to opt in to automatic fallbacks to a lower-tier model when safeguards mandate that Opus 5 declines a user’s request.

Sam Altman didn’t need another lawsuit

2026-07-14T10:03:06-04:00

OpenAI has spent the better part of the year involved in lawsuit after lawsuit, including one from the world’s richest man. But last Friday, the company was hit with one of the highest-profile legal actions yet — from Apple. OpenAI’s expensive hardware bet is what’s on the line.

Apple’s lawsuit against OpenAI, filed in Northern California federal court, accused former Apple employees of “stealing Apple’s trade secrets for the benefit of OpenAI.” The 41-page complaint states that Apple keeps its “product development, manufacturing, supply chain, technology research, and other innovations confidential” and that “the trade secrets spanning Apple’s hardware operations collectively constitute one of the most valuable intellectual assets in all of American business.”

The allegations referenced three former Apple employees: Tang Tan, former VP of Apple Watch, who spent 24 years at Apple before becoming chief hardware officer of OpenAI (after OpenAI bought Jony Ive’s hardware company, io); Chang Liu, former systems electrical engineer for iPhone, who spent eight years at Apple before departing for OpenAI; and Yu-Ting “Alyssa” Peng, another former Apple employee who also left for OpenAI. (Less is known about Peng’s Apple tenure.) The lawsuit makes a lot of wild claims, including Tan allegedly requesting prospective employees to bring Apple hardware outside the office for “show and tell” during their OpenAI interviews and coaching Apple employees about how to avoid the company’s offboarding security procedures. (It’s unclear which, if any, of Apple’s hardware-related trade secrets may have made it into OpenAI’s device, as it does not yet have a publicly shown product.)

Avery Williams, cochair of the trade secret practice area at McKool Smith and author of the firm’s AI Litigation Tracker, told The Verge that he thinks OpenAI’s legal troubles are far from over. “They’ve gotten sued a lot. [But] OpenAI is not going to be out of the woods until we get a ruling from a higher court on the fair use question for AI training. It’s a trillion-dollar question.”

The lawsuit is yet another obstacle for OpenAI, which has had a rollercoaster six months full of drama with rivals over US military red lines, protests over contract with the US government, a drawn-out legal battle with former OpenAI co-founder Elon Musk, and a race with Anthropic over which frontier lab would be first to IPO — not to mention a laundry list of lawsuits, a few of which came from Musk himself. Another still ongoing lawsuit was brought by the family of Adam Raine, a 16-year-old who died by suicide after confiding in ChatGPT. And then there’s the long-running suit brought by The New York Times and other publishers against OpenAI, alleging copyright infringement.

The timing of Apple’s lawsuit is also far from ideal for OpenAI: The company is gearing up to go public, having confidentially submitted a Form S-1 last month with the SEC; it’s facing investor pressure to turn a profit; and it’s cut out a bunch of “side quests” in order to focus on key revenue drivers like enterprise and coding. Perhaps most significantly, though, it’s getting ready to release a much-hyped hardware device in 2027, after paying nearly $6.5 billion to buy famed Apple designer Jony Ive’s hardware startup, io (and handling yet another lawsuit from a company called Iyo). Apple may have a reputation for falling behind in AI, but it’s always had a head start in hardware, and OpenAI has the opposite reputation. A common industry refrain states that “hardware is hard”; for evidence, just take a look at the graveyard of erstwhile overhyped AI devices, like the Humane AI pin.

“The next frontier is going to be AI hardware, more than just chips … Robotics and other types of physical AI is the next area that is ripe for disruption,” said Charlyn Ho, CEO and founder of Rikka Law Group, a law firm focusing on tech, privacy, and cybersecurity law. “It’s kind of interesting to see OpenAI … going into that play, because maybe they’re seeing [that] just the pure software play is not profitable.” She added, “They’re a frontier lab. They’re spending more money than they’re making at the moment.”

It makes sense that OpenAI would want to hire to fill in those hardware gaps and speed up its hardware progress. But now, it’s led to what will likely be a costly legal battle.

“It’s never fantastic to get sued by Apple when you’re trying to IPO,” said McKool Smith’s Williams. “Apple is a tenacious litigant … They do not tend to back down.”

OpenAI did not respond to a request for comment.

Over the weekend, one X user wrote, “Sam Altman wasn’t afraid of Elon but he is terrified of Apple. You can tell by all his posting today.” Altman replied, “i am not afraid of apple, but i have tremendous respect for them. s-tier company.”

Apple and OpenAI have in the past had a relatively positive public relationship; the two in 2024 entered into an agreement to integrate ChatGPT into Apple devices. This lawsuit represents a significant departure. “Apple and OpenAI have been on the same side … As the industry gets more and more competitive, you’re starting to see these alliances fracture,” said Rikka Law Group’s Ho.

In today’s hyper-competitive AI industry, public accusations or lawsuits over trade secrets, corporate espionage, and illicit information harvesting are incredibly common. Point to any AI leader and chances are they’ve got at least one lawsuit or accusation to their name. As far as trade secrets, there’s Scale AI accusing an employee who left for Mercor, xAI accusing employees who left for OpenAI, and Tesla accusing an employee who allegedly transferred data about an AI supercomputer to his personal laptop (and another employee related to robotics-related trade secrets). As far as distillation and improper information harvesting, there’s Anthropic’s accusations to Chinese AI companies and xAI’s alleged distillation of OpenAI’s models.

And although Apple’s lawsuit against OpenAI made headlines, most experts The Verge spoke with said none of the allegations were particularly unusual in and of themselves — it was more interesting that they all happened in the same case and that they were two large players. Trade secrets lawsuits can also be tough to prove in court, since unlike copyright or trademark lawsuits, you’re not typically directly comparing one thing against another. Haibing Lu, professor of information systems and analytics at Santa Clara University’s Leavey School of Business, said Apple’s allegations are “not unique” and that “it’s quite common in Silicon Valley.” McKool Smith’s Williams said that the case reads like a “run-of-the-mill trade secrets misappropriations claim.”

“This isn’t really an AI case as it is a case against an AI company,” Williams continued. “The players are huge, but the allegations are otherwise not unusual.”

For Michael Barnhart, a nation-state threat researcher at cybersecurity firm DTEX, “the layered approach” of the allegations made Apple’s lawsuit interesting: hiring-stage extraction, interview-based solicitation, internal collusion. It’s “very run-of-the-mill in terms of insider threats and this is how they act,” he said. But at the same time, he said, “We can sit there and I’ll pull up a company that did one of these things badly … You’ll have one, you’ll have two, you’ll have three — this one’s got all of them. This one has so many. And also, too, at the highest level.”

It’s somewhat ironic that, in an industry that’s only possible because of the untold amounts of data it’s consumed — whether technically above-board or unauthorized — AI labs get particularly litigious when others take that information elsewhere. After all, the AI industry is a small world, and it’s famously cliquish. Anthropic was founded by ex-OpenAI executives, xAI was founded by an ex-OpenAI cofounder, and countless startups have been acquired (or their core talent acqui-hired) by either the workplaces they were born out of or those companies’ chief competitors. Many of the most influential people at every leading AI company once worked extremely closely with their rivals at another, and in the future, many of them probably will again.

But mentioning that type of irony about the AI industry means you have to bear out that same point across recent decades. Yes, Anthropic allegedly scanned millions of copyrighted books to train its AI models, but the pre-generative AI tech industry spent decades entrenched in similar controversies: Google scanning millions of books to create a searchable database, for instance, or the sheer amount of piracy YouTube indirectly enabled in the early 2000s.

Apple implied in its legal filing that there were significantly more former employees who left for OpenAI with key information than it named in the lawsuit. The company said it had “uncovered a pattern of theft of Apple’s trade secrets by OpenAI employees who were formerly at Apple” and that “this is the tip of the iceberg … At every level, from members of its Technical Staff to its Chief Hardware Officer, and in coordination with business partners, OpenAI has been stealing Apple’s trade secrets and confidential information.”

Alex Terepka, a founding partner of Watstein Terepka LLP, said the case is likely to be contentious and drawn out, potentially over the course of years. “Apple and OpenAI are going to be in this for the long haul,” he said. “What Apple’s put together here is frankly way more than enough to survive a motion to dismiss. Then you’re going to get into discovery, which in these types of cases can be super involved.”

The software side of the AI industry has generated inestimable hype, competition, and controversy, but hardware is the next area AI companies have their sights set on conquering. But since AI devices haven’t truly been tested yet on a broad scale, it’ll take time to see what success in that space looks like — especially with a potential yearslong legal battle between Apple and OpenAI ahead.

Fidji Simo steps down from leading OpenAI’s AGI work due to illness

2026-07-09T19:24:04-04:00

OpenAI’s Fidji Simo is departing her full-time role as the company’s AGI chief and is transitioning to being a “part-time advisor,” she said on X.

The news follows Simo’s original announcement in April that she would take a few weeks of medical leave due to a neuroimmune condition, shortly after she had taken on the AGI chief title. (She had formerly been the company’s CEO of applications). Around the same time, COO Brad Lightcap also stepped down from his role to focus on “special projects,” and OpenAI CMO Kate Rouch also stepped down in order to focus on her health. Rouch planned to return to a “more narrowly scoped role” when her health allowed for it, Simo said at the time.

Simo’s announcement in April also kicked off months of announcements of C-suite changes. While Simo was out on medical leave, OpenAI president Greg Brockman was to take charge of product, including leading OpenAI’s super app efforts, while CSO Jason Kwon, CFO Sarah Friar, and CRO Denise Dresser led the business side of things.

But about a month later, in mid-May, things changed again, with a company reorganization: Brockman officially took charge of OpenAI’s product strategy and “scaling,” leading four different pillars: core product and platform; critical enterprise industries; consumer (i.e., health, commerce, and personal finance); and core infrastructure, ads, data science, and growth. In a memo viewed by The Verge in May, Brockman wrote that the reorganization would help the company prioritize its AI agent goals by combining its products to “invest in a single agentic platform and to merge ChatGPT and Codex into one unified agentic experience for all.”

“Three months ago, I had to go on medical leave after a severe exacerbation of a chronic illness I’ve lived with for seven years,” Simo wrote on X. “During that time, it became clear that the road to recovery would be much longer and more complex than I had anticipated—and that I needed to focus on it fully … It has been a jarring experience to spend my days helping build the future while simultaneously navigating a disabling disease that still has no cure.”

OpenAI CEO Sam Altman wrote on X that he was “really sad about this” and grateful to Simo.

OpenAI rolls out GPT-5.6 after government greenlight — and announces ‘ChatGPT Work’

2026-07-09T13:32:30-04:00

About two weeks after OpenAI’s GPT-5.6 was caught up in regulatory drama — rolled out only to government-approved organizations during a “limited preview” period — the company has received the Trump administration’s greenlight for a public rollout of the model. OpenAI CEO Sam Altman called it “the best model we have ever produced.”

To celebrate, OpenAI also unveiled a new AI agent on the same day: ChatGPT Work. It’s billed as a combination of ChatGPT and Codex, allowing the everyday non-technical user to take advantage of Codex’s capabilities for non-coding tasks, and it’s powered by the GPT-5.6 model suite (Sol, Terra, and Luna). “It can gather context from the apps, files, and workflows you choose and create finished materials such as documents, spreadsheets, presentations, and web apps,” OpenAI wrote in a blog post, adding that a “unified plugins directory” allows ChatGPT to connect to tools like Slack, Gmail, Google Drive, calendars, and CRMs.

Mac and Windows users worldwide, including free ChatGPT users, should have immediate access to ChatGPT Work and GPT-5.6 via the ChatGPT desktop app. On mobile and the web, Pro, Enterprise, and Edu users will first get access, while Plus and Business users will receive access “over the next few days,” OpenAI wrote, adding that the “rollout is starting globally and will continue gradually toward full availability over the next 24 hours.”

Companies like OpenAI and Anthropic, as well as tech giants like Google (and even, recently, Apple), have been vying to clear new ground in the race to make AI agents actually useful for the average person, especially in the wake of the viral open-source AI agent OpenClaw. They’ve had varying results, and for now, the theoretical right-hand AI agent for the everyday consumer remains out of reach.

OpenAI is hoping that its new product, which is a direct competitor to Anthropic’s Claude Cowork (combining its own Claude and Claude Code), will push it ahead in the race.

OpenAI is especially banking on Sol, the most powerful of the GPT-5.6 model suite, to set “a new standard for intelligence and efficiency,” particularly when it comes to coding, cybersecurity, and science, as well as computer use capabilities. The company is also marketing the model as a lower-cost alternative to competitors’ most powerful models, amid complaints of an industry-wide money squeeze and AI lab costs being passed onto customers.

Meta says its new AI model is ready to compete on coding

2026-07-09T10:05:59-04:00

After reentering the AI race with its first in-house Muse Spark model in April, Meta is now opening up the doors to developers with a new model that can plug into AI coding software with the new Meta Model API.

Meta says that Muse Spark 1.1 is a “step-change” from the first generation, with improvements based on feedback from developers. The company says it’s capable of more advanced coding, including detection and fixing of complex bugs; better supports end-to-end agentic workflows across a range of apps, including multi-agent systems; and has native multimodal perception across images, videos, and documents.

The Muse Spark 1.1 launch follows this week’s launch of Muse Image, an image generation model that’s proved controversial for its ability to incorporate other users’ Instagram content into its generations. It’s part of Meta’s race to justify the billions it’s spent on catching up in the AI competition and attempting to achieve parity with companies like OpenAI, Google, and Anthropic, especially after a slew of high-profile hires and a company restructuring last year.

The new 1.1 model is available now in Thinking mode through the Meta AI app and Meta AI website. It will also be accessible through a new Meta Model API, available from today in public preview for US developers. Meta is including $20 worth of free credits with every new Meta Model API account. Muse Spark was initially only available directly through Meta AI, before eventually powering the chatbots inside Instagram and WhatsApp, and the latest Meta smart glasses.

Say hello to Claude Wrapped

2026-07-09T09:40:10-04:00

The popularity of Spotify Wrapped has kicked off a wide range of year-in-review features, on apps from YouTube to Uber — and now, the lookback trend has come to AI. Anthropic on Thursday announced a “reflect” feature for its Claude chatbot, allowing users to see an analysis of their usage data over the past month, three months, six months, or year.

Anthropic bills the reflection dashboard as a way to “see your patterns and shape them,” the company wrote in a blog post. It begins with a summary of an individual’s key topics brought up with Claude, as well as types of tasks they delegate and their usage patterns, including peak usage times. Users will also be able to set their own “quiet hours” or break reminders after certain amounts of time. (Anthropic also said that an analysis of total time spent using Claude, seeming to imply a metric like hours of usage tallied up, would be coming soon.)

“It lets you easily track and visualize how you use Claude, and decide whether that time aligns with your goals … Your reflection also invites you to step back and examine the role Claude plays in your life. It will periodically surface questions like, ‘What’s one thing you want to keep doing yourself, even if Claude could do it faster?’” Anthropic wrote. Somewhat ironically, once a user answers such a question, the company said the dashboard would then give them “the chance to talk it through with Claude.”

Anthropic has spent a lot of money marketing itself as the company behind an “AI collaborator” that allows people to think more deeply — there have been ads, billboards, and even literal “thinking” caps. And Claude’s “reflect” feature is billed the same way, with the company’s blog post instructing people to use it to “build AI skills that support your original thinking.” The feature will show examples of how you interact with Claude, “like noting that you often rework email drafts in your own voice, or delegate tasks only after settling the strategy yourself,” Anthropic wrote.

Anthropic said the reflection dashboard won’t pull from files within connected tools or platforms (like the actual emails within a connected email account), though the fact that a user may have used Claude to summarize their email inbox may come up in the reflection. It also won’t pull from chats that occurred in incognito mode or in any way reference a “conversation connected to a health integration tool.” As far as “sensitive topics,” which Anthropic is vague about defining, the company wrote that “sensitive conversations can still appear as part of your reflection, but only at a high level.”

The feature, which is currently available in beta mode for free users of Claude, as well as Pro and Max subscribers, will work for those who have memory turned on for their chats, and it’s accessible via Settings on either the web or the Claude desktop app. The idea came from interviews with Claude users, Anthropic said, and the feature will also come to Claude Cowork “soon.”

Anthropic’s long-sidelined Fable 5 is greenlit to return

2026-07-01T09:11:31-04:00

After weeks of negotiating with the Trump administration, Anthropic is finally going to be able to bring Claude Fable 5 back online. In a post on X, Anthropic said it plans to begin restoring access Wednesday to users globally on Claude platforms, and that the company would re-enable access on AWS, Google Cloud, and Microsoft Foundry soon, but with no set timeline.

Anthropic wrote:

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.

We’ll begin restoring access tomorrow, and will share an update soon.

We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.

The company also published a blog post on Tuesday evening detailing the lead-up to the events in question, its revised safeguards, new AI industry processes it’s working on, and new ways it plans to share information with the government, as well as prerelease testing for upcoming models.

In early June, Anthropic sidelined Fable 5 — its consumer-facing model built from the same underlying technology as Anthropic’s Mythos 5, but with more safeguards — after a Friday evening ultimatum from the Trump administration. Due to concerns over potential jailbreaks of the technology, the government had slapped Anthropic with an export control directive, disallowing any foreign national (including non-US members of enterprise client companies and even many of Anthropic’s own employees) from using either Mythos 5 or Fable 5, i.e., both the models that Anthropic had spent the last week hyping.

Due to concerns over potential jailbreaks, the government slapped Anthropic with an export control directive, affecting both the models that Anthropic had spent the last week hyping

To address the jailbreak in question, which Amazon researchers had flagged and was largely responsible for setting the export control directive into motion, Anthropic said in the blog post that it had “trained an improved safety classifier that targets and blocks” that behavior, adding, “Users will be notified if a request to Fable 5 is blocked, and the request will instead be sent to Opus 4.8. The new classifier means that the specific technique described in the Amazon report is blocked in over 99% of cases.”

The Trump administration recently greenlit the return of Mythos 5, but to a preapproved list of organizations only. Non-US members of those organizations, as well as Anthropic’s own foreign national employees, were allowed to regain access to the model. The decision came shortly after OpenAI unveiled GPT-5.6, which the Trump administration only allowed to debut under similar rules: a staggered rollout, initially to only a preapproved list of organizations and government departments.

On Tuesday, Anthropic wrote that the company will “continue to coordinate with the government to expand access to the broader set of domestic and international partners” for Mythos 5. The company also included an entire section in its blog post on its new plan for working closely with the Trump administration — highlighting its many attempts to get back in the government’s good graces after months of public drama, lawsuits, and presidential action.

The company wrote that it plans to offer “pre‑release government access and evaluation,” particularly for models relevant to national security capabilities, so that government partners can run independent evaluations about the models’ capabilities and test guardrails before broader release. (The government will also have access to Anthropic’s technical staff during those prerelease testing periods.) Anthropic said it also plans to introduce “rapid information sharing” when “significant jailbreaks or misuse patterns are identified.”

It said it would work with the government and other leading AI labs to create a “shared, voluntary security and evaluation standard for frontier model providers.” Finally, Anthropic said it would “stand up dedicated Anthropic teams to work on shared government priorities, provide a significant compute allocation to support government testing and research, and make our safety and red‑teaming expertise available to help advance the state of the art in AI evaluation.”

Anthropic said it would work with the government and other leading AI labs to create a “voluntary” shared standard for security and evaluation.

The initial export control directive by the Trump administration came at an inopportune time for Anthropic, as it’s preparing for an IPO and has been feuding with the government for months over a supply chain risk designation.

The company highlighted in the blog post that there’s “currently no consensus in the AI industry” for deciding on a jailbreak’s severity, a problem that “will become more acute in the coming months, as more models with powerful cybersecurity (and other) capabilities are trained, assessed, and released.” So it said it partnered with Amazon, Microsoft, Google, and other enterprises that are part of its Project Glasswing program in order to draft a widely agreed upon framework for assessing AI jailbreaks, with four proposed categories: capability gain for the attacker, breadth of capability gain for the attacker, ease of weaponization more broadly, and discoverability (or how easy it may be for someone else to repeat it). Anthropic said it had also created a new team to “provide 24/7 monitoring of key jailbreak submission channels” and would also soon debut a HackerOne program for researchers to submit potential jailbreaks they have flagged for Fable 5.

Anthropic also included a disclaimer in its blog post, writing that “it is probably impossible to make any AI model fully robust (that is, impervious) to jailbreaks. We expect that some jailbreaks will be found for our models, and that they will vary in severity: there will be many minor jailbreaks, some narrow harmful ones, and although no universal jailbreaks for Fable 5 have been discovered at the time of writing, expert safety researchers continue to red-team it.”

Lawmakers want to ban AI companies from selling your health data

2026-06-29T12:07:14-04:00

A new proposal would ban the sale of Americans’ health and location information to data brokers — including information people reveal to an AI chatbot like ChatGPT or Claude.

In the coming weeks, Senator Elizabeth Warren (D-MA) and Representative Mary Gay Scanlon (D-PA) are planning to debut a new version of the Health and Location Data Protection Act that’s better suited to the AI era. The former version of the bill, first introduced in June 2022, prohibited data brokers from collecting and selling health and location data. Four years later, it’s expanded to ban other companies from selling such data to brokers, and to specifically cover data entered into AI systems.

AI labs have set their sights on building health and medical products. In January, Elon Musk publicly called for people to upload their medical records, like MRI scans, to Grok, xAI’s chatbot. That same month, OpenAI introduced ChatGPT Health, a sandboxed tab within ChatGPT that it deemed more secure, and encouraged users to upload their medical records and other sensitive information. It also introduced ChatGPT for Healthcare, aimed at medical providers. A few days after that, Anthropic quickly followed up with Claude for Healthcare, a “HIPAA-ready” tool for individuals, health providers, and hospitals.

But when battling data breaches or unauthorized access to data, users are largely at the AI companies’ mercy. The data protection for tools like OpenAI’s and Anthropic’s “largely depends on what companies promise in their privacy policies and terms of use,” Sara Gerke, a law professor at the University of Illinois Urbana-Champaign, told The Verge in January. The US lacks an overall federal framework for data privacy, despite years of attempts.

This bill — which is also sponsored by Senators Ron Wyden (D-OR) and Bernie Sanders (I-VT) — would require the Federal Trade Commission to enact the rules within 180 days and will allow the FTC, state attorneys general, and affected individuals to sue to enforce it. It would also earmark $1 billion to the FTC over the next 10 years for enforcement.

“It’s more important than ever that we crack down on data brokers that are raking in giant profits from selling Americans’ most sensitive information,” said Senator Warren in a statement. “Especially as more people enter their private health data into AI, we need to make sure that information isn’t exploited by the highest bidder.”

Anthropic’s Mythos 5 is back

2026-06-29T08:13:00-04:00

After a rollercoaster negotiation process with the Trump administration that dragged on for two weeks, Anthropic’s Mythos 5 is finally back in action — at least, somewhat, for a select group of organizations, according to a letter from the government to Anthropic that was viewed by The Verge. Fable 5, however — the public-facing Mythos-class model — appears to still be in limbo, with no apparent timeline for a rollout agreement.

The letter, dated June 26th and sent by Commerce Secretary Howard Lutnick to Anthropic cofounder Tom Brown, who had been recently leading negotiations, states that there’s been a “revision to the license requirements” based on the fact that Anthropic has recently “worked with the U.S. government to address risks” associated with Mythos 5 and Fable 5.

Anthropic spokesperson Danielle Ghiglieri told The Verge in a statement that the company had “received notice from the US government that Mythos 5, our strongest cybersecurity model, can be redeployed to a small group of cyber defenders and infrastructure providers.” She added, “We are working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible. We are pleased to see this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.”

The US government did not lift the export control directive it hit Anthropic with two weeks ago, which barred any foreign national from accessing either model (including Anthropic’s own employees). Instead, the government essentially made an exception for Mythos 5, approving a select group of organizations for access the same way it did for OpenAI’s GPT-5.6, announced earlier today. Under this exception, Anthropic employees who aren’t US nationals and members of the approved organizations who aren’t US nationals are all greenlit to access Mythos 5, according to the letter.

“These efforts have yielded significant progress,” Lutnick wrote. “In addition, Anthropic has committed to work with the U.S. government on protocols and standards and releases for [Mythos-class models]. In light of this progress … I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.”

Pressure has been building on the Trump administration to make a change in the case-by-case type of regulatory environment it recently adopted, especially as Anthropic’s competitors’ cybersecurity-focused models kept getting better and better — and even pulling ahead of it on some cybersecurity-focused benchmarks. Pressure was also building within the US AI industry, particularly fears about the AI advancements China could make while top US AI labs were sidelined. Plus, there was the fact that top US government departments, like the National Security Agency, had lost access to Mythos 5.

Now, Anthropic has essentially the same deal OpenAI has: Release the model in a limited preview setting, for approved organizations only, such as trusted enterprises and the US government itself. Both AI labs hope that more general availability comes soon, both for enterprise deals and for public access (like with Anthropic’s Fable 5), but it’ll be ultimately up to the Trump administration. Though many have called for AI regulation, including AI labs themselves, a current consensus among some tech leaders is that the past two weeks haven’t been the way to do it.

In fact, OpenAI wrote in the company’s GPT-5.6 blog post, “We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.”

Lutnick wrote in his letter to Anthropic, “All other requirements of the June 12 letter remain in effect until further notice.” He added, “I reserve the right to reevaluate and adjust the scope of license requirements on [Mythos 5 and Fable 5], should circumstances change.”