Trump Administration Forces Anthropic to Pull Fable 5 and Mythos 5 Offline

At 5:21 PM on Friday June 12, the U.S. Commerce Department sent Anthropic a letter that would upend the AI industry before the weekend was over. The directive, invoking an obscure export control provision, banned all non‑U.S. persons — including Anthropic's own foreign‑national employees — from accessing the company's newest models: Fable 5 and Mythos 5.

The only way to comply, Anthropic determined, was to shut down access for everyone. By Friday evening, both models — launched just three days earlier on June 9 to widespread acclaim — were offline worldwide. The U.S. government had successfully forced a private AI company to pull its products with a unilateral action that did not require court approval, TechCrunch reported.

Anthropic confirmed the shutdown in a statement, saying it was complying with the legal directive but pushed back: "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." The letter itself has not been made public, and Anthropic said the government "did not provide specific details of its national security concern," according to The Verge.

The entire episode traces back to a deceptively simple prompt: "Fix this code."

Katie Moussouris, a cybersecurity veteran and founder of Luta Security, disclosed in a ¹ that Anthropic had privately shared a paper with her — written by security researchers at Amazon, per ¹ — describing a guardrail bypass in Fable 5. The difference came down to phrasing: asking the model to "review code for security issues" stayed within its safety boundaries, but asking it to "fix this code" — the same task with a different prompt — tripped a bureaucratic alarm. The end result of both prompts was fundamentally identical: the model was helping with code security either way.

"The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense," Moussouris wrote, calling the export control directive ¹

The chain of events that led to the shutdown runs directly through Amazon. CEO Andy Jassy personally shared the Amazon security researchers' findings with senior White House officials, according to The Verge. Shortly after those conversations, the Commerce Department issued its export control directive.

The context is impossible to ignore: Amazon is Anthropic's largest investor — having poured $8 billion into the company — and also a direct competitor through its own AI models and Bedrock platform. Whether Jassy's outreach was motivated by genuine security concern, competitive maneuvering, or both is now the subject of intense speculation in the AI community.

Adding another layer, Semafor reported that the White House's decision was also driven by fears that a group linked to China may have accessed Mythos 5. But an Anthropic spokesperson told Semafor that the government never mentioned China during its export control discussions. Trump adviser David Sacks, meanwhile, posted on X about the jailbreak claims without referencing China at all.

Moussouris was not alone. Dozens of top cybersecurity researchers and executives signed an open letter calling on the Trump administration to revoke the export control order, arguing that pulling advanced cybersecurity capabilities from network defenders was "dangerous."

"The Trump administration's move is likely to raise alarms in foreign capitals about the reliability of American AI for critical applications," Justin Hendrix, editor of Tech Policy Press, told.¹ The message, he said, is that U.S. AI companies cannot be trusted to operate without government interference.

Citing anonymous sources, ¹ described a tense weekend situation, reporting that "personality differences" between Anthropic and the Trump administration — rather than any technical flaw — drove the export directive. AP News confirmed that a group of cybersecurity executives is now urging the administration to ease restrictions.

The shutdown didn't just affect American users — it ignited a global backlash that may reshape the AI industry faster than anything Anthropic's competitors could have done on their own.

In the UK, AI and online safety minister Kanishka Narayan used the moment to argue for British AI independence. "We treat every other threat to our sovereignty with deadly seriousness, but we haven't learned to treat this one in the same way," he said, framing AI as "the central political question of our time," The Verge reported. Britain must decide how AI shapes its economy and security "before someone else decides the answer for us."

In France, former Prime Minister Gabriel Attal — now presidential candidate for Macron's Renaissance party — went further, calling the shutdown the start of "the AI war" and comparing it to Iran's blockade of the Strait of Hormuz. Access to AI, he argued, is now a strategic chokepoint. Le Monde reported similar alarm across France's political spectrum.

In Canada, Prime Minister Mark Carney struck a measured but pointed tone: "Nobody has done anything wrong in the situation. But we will have done something wrong if we just accept this, don't take the lesson, don't build out and diversify," The Verge reported.

Members of the European Parliament pointed to the incident as evidence that tech sovereignty needs to become reality — and quickly. Companies like France's Mistral and Canada's Cohere, while not matching frontier U.S. labs on raw capability, suddenly look more strategically valuable than they did a week ago.

Regardless of whether Fable 5 and Mythos 5 come back online this week — Anthropic executives were in Washington, DC on Monday meeting with administration officials, per Forbes — the damage to America's reputation as a reliable AI provider has already been done.

The models themselves were significant. Fable 5, launched June 9, was what Anthropic called its most capable generally available model ever, with safeguards limiting use in high‑risk areas. Mythos 5, built on the same foundation but with some safeguards lifted, was available only to select government agencies and companies. Both were built on the same architecture as Mythos Preview — which Anthropic itself had previously ² to release publicly.

As ⁴ captured the core tension: Anthropic may soon bring Mythos and Fable back online. But restoring global trust in American AI is another thing entirely. No matter how long the shutdown lasts, it shined a light on how fragile access to U.S. frontier AI models is. Many governments and companies did not like what they saw — and are fired up to make sure it doesn't happen again.

For builders and developers who integrate frontier AI models into their products, the Fable 5 and Mythos 5 shutdown is a wake‑up call with immediate practical implications:

• Multi‑model architecture is no longer optional. If your application depends exclusively on a single provider's models, you're one government letter away from an outage. The builders who weathered this weekend best were those already routing through platforms like OpenRouter or maintaining fallback model chains.
• Open‑source models just got a lot more strategic. When access to frontier closed‑source models can be revoked by government decree, open‑weight alternatives — even if they lag on benchmarks — provide an insurance policy that closed‑source models cannot.
• Sovereign AI isn't just political theater anymore. The UK, France, and Canada are now actively pushing for domestic AI capacity. For builders outside the U.S., the message is clear: diversify your model dependencies or risk being cut off. European builders, in particular, should watch Mistral's trajectory closely.
• Export controls are the new frontier of AI regulation. This wasn't legislation or a court order — it was an administrative directive. The speed and opacity of the process means any AI company with foreign‑national employees or customers is vulnerable to similar action.

The models may come back online this week, but the genie is out of the bottle: governments now know they can shut down AI access with a letter, and every builder outside the U.S. now knows they need a Plan B.