5-Min Brief: The US Government Just Forced Anthropic to Shut Down Its Two Most Powerful AI Models. Here's the Full Story.

What you need to know — in 30 seconds

• On June 12, 2026 — three days after launch — the US government ordered Anthropic to shut down Claude Fable 5 and Mythos 5 worldwide
• The stated reason: the government believes it found a method to jailbreak Fable 5 into identifying software vulnerabilities in code — a national security concern
• Anthropic complied, taking both models offline for every user on the planet, but publicly disputes the government's reasoning
• All other Claude models — Opus 4.8, Sonnet 4.6, Haiku 4.5 — remain fully available
• This is the first time a US export-control directive has ever been used to pull a publicly deployed AI model

Full disclosure: Claude — the AI that helps produce this newsletter — is made by Anthropic. This is the biggest story involving Anthropic since we launched, and I want to cover it as straight as I can. That means explaining both what the government did and why Anthropic publicly disagrees with it.

Here's the full story.

What happened — timeline

June 9, 2026. Anthropic launched Claude Fable 5 — its first public release from the Mythos-class family of models, which sits above Opus in Anthropic's internal model tier. Fable 5 was the most capable model Anthropic had ever made available to the public. Mythos 5 — even more powerful — was simultaneously made available to a small set of authorized users.

March 2026 — the background. This launch happened against an existing backdrop of tension. In early March, the Department of Defense had classified Anthropic as a "supply chain risk" — a significant legal designation under 10 U.S.C. § 3252. Dario Amodei publicly called the classification legally untenable and said Anthropic planned to challenge it in court. According to Axios, the Trump administration had actually tried to stop Anthropic from releasing Fable 5 entirely and failed. The release happened anyway.

June 12, 2026, 5:21pm ET. Commerce Secretary Howard Lutnick sent a letter to Dario Amodei invoking US export control authorities. The directive: suspend access to Claude Fable 5 and Mythos 5 for any foreign national — inside or outside the United States, including Anthropic's own foreign-national employees.

The problem with that order is that Anthropic cannot filter foreign nationals from US users in real time. There's no mechanism to do it instantly and reliably. So Anthropic made the only decision it could: it shut both models down for everyone, everywhere, to ensure compliance.

By 9:59pm ET that night, access was gone.

What the government said — and what Anthropic says back

The government's stated reason for the shutdown is a reported jailbreak of Fable 5. Specifically, the government claims someone found a method to instruct Fable 5 to read a specific codebase and identify software vulnerabilities — turning the model into a tool for finding security holes.

That should sound familiar to readers who've been following this newsletter. Back in April, we covered Anthropic's Project Glasswing — their AI system that found thousands of security holes in major software that humans had missed for decades. The concern at the time was exactly this: that a sufficiently capable model could be used to find and exploit software vulnerabilities. The government is now saying that concern is real and present with Fable 5.

Anthropic's response has been unusually direct and public. In its official statement, the company said it had received only verbal evidence of a "potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws." Anthropic's position: the capability the government is worried about is already widely available in other models — including OpenAI's GPT-5.5 — and is used every day by defenders who keep systems safe. Applying this standard across the industry would halt all new frontier model launches.

Anthropic is complying while it contests the decision. The company said it would share more technical details within 24 hours of the shutdown. Senior Anthropic technical staff have been in Washington meeting with White House officials to try to resolve the dispute, and both sides have said publicly they're eager to find a resolution.

Why this is different from anything that's happened before

This isn't a company voluntarily pulling a product due to safety concerns. It isn't a court order following litigation. It's the first time in history that a US export control directive has been used to force the immediate worldwide shutdown of a publicly deployed AI model.

The legal mechanism matters. Export control law is designed to prevent sensitive technology from reaching foreign adversaries — it's the same framework used to restrict semiconductor exports to China, for example. Applying it to a publicly available AI model is a significant expansion of how that framework has historically been used.

The scope is equally significant. The directive targeted foreign nationals — not just foreign countries. That includes Anthropic's own employees who hold foreign citizenship and work in the United States. The practical impossibility of filtering for that in real time is why every user on the planet lost access, not just foreign nationals.

The precedent this sets is the part that will matter most long-term. Every AI company now knows that the US government has a tool it's willing to use to pull a live deployed model — not just restrict new releases, but take down something already in customers' hands. How companies design their infrastructure, their user agreements, and their relationships with regulators will look different in light of June 12.

What it means for the AI industry — and for the IPO

The timing here is almost impossible to separate from everything else happening in Anthropic's life right now. The company filed its confidential S-1 with the SEC on June 1 — the first step toward an IPO at a valuation near $1 trillion. We covered that last week.

An ongoing dispute with the US government over its most powerful models is precisely the kind of risk factor that will need to appear in that S-1 when it becomes public. The DoD "supply chain risk" classification, the forced model shutdown, the legal challenge — all of this is material information that investors evaluating a trillion-dollar listing will need to understand.

For enterprise customers who built applications on Fable 5 in its three days of availability: Anthropic's recommended fallback is Opus 4.8, which remained fully online and unaffected. The incident is also prompting broader conversations among enterprise AI teams about what it means to depend on any single model provider — and how to build for resilience when a critical dependency can disappear on a Friday evening without warning.

What happens next

As of this writing, Fable 5 and Mythos 5 remain offline. Anthropic says it is working to restore access and challenging the government's legal basis for the directive. Senior technical staff are in active discussions with the White House.

The most likely outcomes, in rough order of probability: a negotiated restoration with technical modifications to satisfy the government's security concerns; a prolonged legal battle while the models remain offline; or a permanent restriction that changes what Anthropic can offer publicly going forward.

What happens will matter well beyond Anthropic. The question of how much control the US government has over what AI companies can deploy — and to whom — is now being answered in real time, with real consequences for the companies, their customers, and the technology itself.

HumanReadable-AI covers AI news in plain English every weekday. Subscribe below — free, no jargon, always under five minutes.