A narrow jailbreak shouldn't justify recalling models used by hundreds of millions
In the span of a single week, Anthropic launched and then lost two of its most powerful AI models — not to a technical failure, but to a government order. The U.S. Commerce Department, invoking national security authority, directed the company to take Fable 5 and Mythos 5 offline for all users worldwide, citing a jailbreak vulnerability that Anthropic itself disputes as too narrow to justify such a sweeping response. It is the first time a major AI company has been compelled by direct government intervention to pull a publicly deployed model from the world's hands, and it arrives at a moment when the line between innovation and national security has never been more contested.
- Within days of their celebrated launch, Fable 5 and Mythos 5 were forced offline by a Commerce Department directive — a first in the history of commercial AI deployment.
- The government's alarm centers on a jailbreak technique that could let users bypass Fable 5's safety guardrails, but Anthropic says only verbal evidence was offered and that similar flaws likely exist across the industry.
- Anthropic is pushing back publicly, arguing that a narrow vulnerability does not justify recalling a model used by hundreds of millions of people and that the action fails to meet standards of transparency or technical rigor.
- The shutdown deepens an already fractured relationship: the Trump administration previously barred Anthropic from federal agencies, Trump called its leadership 'leftwing nut jobs,' and the company is currently suing the administration over free speech violations.
- Paradoxically, the NSA was reportedly using Anthropic's Mythos model for offensive cyber operations just days before the shutdown — suggesting the government's posture toward the company is less principled opposition than volatile improvisation.
On a Friday evening in mid-June, Anthropic received an order from the U.S. Commerce Department that left the company little choice: disable Fable 5 and Mythos 5 — its two newest and most capable AI models — for every user on earth. Commerce Secretary Howard Lutnick had invoked national security authorities, pointing to a jailbreak technique that researchers said could allow users to push Fable 5 past its built-in safety limits. The models had launched just days earlier to considerable fanfare. By Friday night, both were gone.
Anthropic did not go quietly. The company acknowledged the vulnerability existed but argued forcefully that the government's response was disproportionate — that only verbal evidence had been provided, that similar weaknesses likely existed in competitors' systems, and that pulling a model deployed to hundreds of millions of people over a narrow flaw was not a reasonable remedy. Fable 5 had been built with especially strict guardrails around sensitive domains like cybersecurity and biology; Mythos 5, a less restricted version, had been made available only to a curated group of trusted infrastructure and security partners.
The forced shutdown is the first of its kind in the commercial AI era, and it lands inside a relationship already defined by conflict. Earlier this year, the Trump administration had barred Anthropic's products from federal agencies after the company sought limits on Pentagon use of its technology. Trump personally attacked the company's leadership on Truth Social. Anthropic sued, and a federal judge ruled in its favor — though the case continues.
There had been tentative signs of reconciliation. The NSA was reportedly using Mythos for offensive cyber operations as recently as last week, and a June 2 executive order on AI had gestured toward voluntary cooperation between the government and leading AI developers. Anthropic's public statement made clear it viewed Friday's directive as a betrayal of that emerging framework — a unilateral move that lacked transparency, technical grounding, and due process. The question of who ultimately governs the deployment of powerful AI systems, and by what rules, remains wide open.
On a Friday evening in mid-June, Anthropic received a directive from the U.S. Commerce Department that forced the company to make an unusual choice: disable two of its most powerful AI models for every single user, not just foreign nationals. The order came from Commerce Secretary Howard Lutnick and cited national security authorities as justification. Anthropic, the maker of the popular Claude chatbot, had released Fable 5 and Mythos 5 just days earlier on Tuesday, marketing them as the most capable systems the company had ever built. By Friday night, both models were offline.
The government's concern centered on a specific vulnerability—a technique that could allow users to circumvent some of the safety guardrails built into Fable 5. The Commerce Department's Bureau of Industry and Security had identified what researchers call a "jailbreak," a workaround that might let someone push the model past its intended boundaries. Anthropic acknowledged the issue existed but pushed back hard on the government's response. The company noted that the government had provided only verbal evidence of the problem and that similar vulnerabilities likely existed in AI systems built by other companies too. "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic wrote in its public statement.
The two models had been designed with different audiences in mind. Fable 5 came with strict safeguards, especially around sensitive topics like cybersecurity and biology, and was released to the general public. Mythos 5, built on the same technical foundation but without those restrictions, went only to a select group of trusted partners—key cybersecurity and infrastructure companies. Anthropic had been explicit about why such caution was necessary: Fable 5's capabilities in areas like cybersecurity could be weaponized if left unguarded. Yet the company believed the government's move was based on a misunderstanding and said it hoped to restore access as soon as possible.
This moment marks the first time a major AI company has been forced to pull a publicly deployed model offline due to direct government intervention. It represents a significant escalation in an already contentious relationship between Anthropic and the Trump administration. In February, the president and Defense Secretary Pete Hegseth had moved to bar Anthropic's products from federal agencies after the company sought stronger restrictions on how the Pentagon could use its technology. Trump had posted on Truth Social that Anthropic's leadership were "Leftwing nut jobs" who had made a "DISASTROUS MISTAKE." Anthropic sued, arguing the administration had violated its freedom of speech. A federal judge in California ruled in Anthropic's favor, though the case remained active in Washington.
There had been signs of a thaw. Last week, the Financial Times reported that the National Security Agency was using Mythos to conduct offensive cyberattacks—suggesting some level of cooperation between the company and the government. On June 2, Trump had signed a landmark executive order on AI that directed federal agencies to strengthen cyber defenses and to design a mechanism for gaining early access to the most powerful AI models from companies like Anthropic and OpenAI on a voluntary basis. The order seemed to open a door for collaboration.
But Anthropic's statement made clear the company saw Friday's directive as a departure from that framework. "We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts," the company wrote. "This action does not adhere to those principles." The tension between national security concerns and the rapid deployment of powerful AI systems remains unresolved, and this forced shutdown suggests the government is willing to act unilaterally when it perceives a threat—even if the company being ordered to act believes the threat has been overstated.
Notable Quotes
We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.— Anthropic, in its statement
We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.— Anthropic, in its statement
The Hearth Conversation Another angle on the story
Why would the government force Anthropic to take the models offline for everyone, not just foreign nationals? That seems like overkill.
Because once you deploy something to the public, you can't surgically remove it from just one group of users. The safeguards are built into the system itself. You'd have to shut it down entirely or rebuild the whole thing with different access controls—which takes time they apparently didn't have.
But Anthropic says the vulnerability exists in other companies' models too. Why single them out?
That's the real question. It could be that Anthropic's models are more widely used, or that the government has a specific reason to worry about this particular jailbreak. Or it could be political—there's a history here.
The February incident with Trump and the Pentagon guardrails?
Exactly. That lawsuit is still active. So when the government says "national security," Anthropic hears something that sounds more like retaliation. The company is arguing the government didn't even show them hard evidence—just told them verbally about the problem.
What does Anthropic want now?
To turn the models back on. They think this is a misunderstanding that can be cleared up. But they're also making a broader point: if the government wants to block AI deployments, there should be a transparent, technical process—not an emergency directive based on a phone call.
And the NSA was already using Mythos for cyberattacks?
Yes. Which makes the timing strange. The government was comfortable enough with the model to use it themselves, but not comfortable enough to let the public have it. That contradiction is at the heart of Anthropic's complaint.