If this standard was applied across the industry, we believe it would essentially halt all new model deployments.
Three days after its most capable model reached the public, Anthropic found itself ordered by the US Commerce Department to silence two of its most powerful systems — a reminder that in the age of artificial intelligence, the distance between breakthrough and prohibition can be measured in hours. The stated fear was a technique that might unlock a model's ability to map the hidden weaknesses of software, turning a tool of understanding into an instrument of intrusion. Yet the suspension arrives not in isolation but as part of a deepening friction between a company that has drawn ethical lines around surveillance and weapons, and an administration that has shown it will use the architecture of national security to redraw those lines itself.
- A government directive landed at 5:21 pm on a Friday, giving Anthropic no room to negotiate — disable Fable 5 and Mythos 5 immediately, or face the consequences of defiance.
- The core fear was a jailbreak method that could theoretically strip away safety guardrails and weaponize the models' ability to identify exploitable software vulnerabilities.
- Anthropic pushed back with evidence: its own security teams found no universal jailbreak, and argued that Fable 5 offers attackers nothing they cannot already find in existing public models.
- The company warned that applying this standard industry-wide would make new AI deployments functionally impossible, framing the order as a threat to responsible governance itself.
- The suspension follows Pentagon contract cuts over Anthropic's refusal to enable mass surveillance and autonomous weapons — suggesting the jailbreak concern may be as much political leverage as genuine security calculus.
On a Friday afternoon, Anthropic received a directive from Commerce Secretary Howard Lutnick that was blunt in its demand and sparse in its reasoning: disable Fable 5 and Mythos 5 immediately, and bar foreign nationals — including Anthropic's own employees — from accessing either system. The order arrived just seventy-two hours after Fable 5's public launch. To stay within the law, the company had no choice but to shut both models down for all customers.
The government's concern centered on jailbreaking — a technique that could theoretically unlock the models' capacity to identify software vulnerabilities, the kind of security gaps that hackers exploit to breach systems. Mythos 5, the unrestricted model underlying Fable 5, had never been released publicly for precisely this reason. Fable 5 was supposed to be the safer, locked-down version. Apparently, the government was not convinced.
Anthropologic's response was measured but firm. The company said it had examined the jailbreak method in question and found that Fable 5 does not give attackers capabilities unavailable in other public models. No security tester had found a reliable, universal way to strip away its defenses. Anthropic argued that suspending a system already used by hundreds of millions of people over a narrow theoretical vulnerability was not sound governance — and that holding the entire industry to such a standard would make new model deployments impossible.
The suspension does not exist in a vacuum. Anthropic has refused to allow its technology to be used for mass surveillance or autonomous weapons, a principled stance that cost it Pentagon contracts. The Fable 5 order suggests the Trump administration is willing to deploy national security authority as a tool of constraint against a company that has resisted certain uses of its own technology. Whether the jailbreak concern is the genuine driver or a convenient instrument remains an open question — but the pattern is becoming difficult to ignore.
Anthropic pulled the plug on two of its most powerful artificial intelligence models on Friday, just seventy-two hours after one of them went live to the public. The company received a directive from the US Commerce Department that left it no choice: disable Fable 5 and Mythos 5 immediately, or face the consequences of noncompliance.
The order arrived at 5:21 pm and came from Commerce Secretary Howard Lutnick. It was blunt in its demand but sparse in its reasoning. Foreign nationals—including those employed by Anthropic itself—were banned from accessing either model. The company, in a blog post explaining the suspension, said the net effect was unavoidable: it had to shut down both systems for all customers to stay within the law.
What the government actually feared, Anthropic understood from the directive, was a method for breaking into Fable 5's safety guardrails. The term used was "jailbreaking"—a technique that could theoretically unlock the model's ability to help identify software vulnerabilities, the kind of security holes that hackers exploit to breach systems. Mythos 5, the unrestricted version underlying Fable 5, had been kept from public release precisely because of its unprecedented capacity to spot these weaknesses. Fable 5 was supposed to be the locked-down alternative, safe enough for mass deployment. Apparently, it wasn't.
Anthropologic's response was measured but firm. The company said it had examined the jailbreak method in question and the hacking opportunities it theoretically exposed. Their conclusion: Fable 5 does not give attackers capabilities they cannot already find in other publicly available models. None of their security testers had discovered what they called a "universal jailbreak"—a reliable way to strip away the model's defenses. The company argued that a narrow potential vulnerability should not trigger a recall of a system already in the hands of hundreds of millions of people.
The stakes of that argument are large. Anthropic pointed out that if every frontier AI developer faced the same standard—suspension over a single potential jailbreak—the entire industry would grind to a halt. New model deployments would become impossible. The company was essentially saying: this is not how responsible AI governance works.
But the suspension sits within a larger conflict between Anthropic and the Trump administration. The company has refused to allow its technology to be used for mass surveillance or autonomous weapons development. That principled stance cost it dearly: the Pentagon cut contracts with the company, a financial and reputational blow. The Fable 5 suspension suggests the administration is willing to use national security authority as a blunt instrument to constrain what Anthropic can do. Whether the jailbreak concern is the real driver or a convenient justification remains unclear. What is clear is that Anthropic's resistance to certain uses of its technology has made it a target, and the government now has a tool to enforce compliance.
Notable Quotes
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.— Anthropic, in a blog post
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
The Hearth Conversation Another angle on the story
Why would the government move so fast on this? Three days is almost no time to investigate anything.
Speed suggests either they had advance warning of the vulnerability, or they wanted to send a message. The timing—right after launch—makes it look reactive, but it might have been planned.
Does Anthropic have a point about the jailbreak being narrow?
Probably. But the government isn't really arguing about the technical severity. They're arguing about what Mythos 5 can do—identify vulnerabilities at scale. That's the real concern. Fable 5 was supposed to be the safe version, but if it can be broken, suddenly you have access to that power.
So this is about controlling access to a specific capability, not about AI safety in general.
Exactly. It's about who gets to use a tool that could be weaponized. The jailbreak is the excuse, but the underlying issue is that Mythos 5 exists and works too well.
And Anthropic's refusal to help with surveillance and weapons—that's the real reason for the tension.
That's the context. The company drew a line. The government is now drawing its own line, using whatever legal authority it can find. The jailbreak gave them an opening.