Three days. That’s how long Anthropic’s most powerful AI model survived before the US government pulled the plug.

On Friday the 13th — because of course it was — the Trump administration issued an export control directive ordering Anthropic to suspend all access to Claude Fable 5 and Mythos 5 for any foreign national. Since there’s no clean way to verify every user’s nationality in real time, Anthropic had to shut down both models for everyone.

This isn’t a regulatory hiccup. This is the US government reaching into a private company and flipping the off switch on its flagship product.

What Got Killed and Why It Matters

Fable 5 is Anthropic’s most capable publicly available model — a “Mythos-class” system above the previous Opus tier. Its restricted twin, Mythos 5, is what Anthropic describes as having “the strongest cybersecurity capabilities of any model in the world,” deployed through Project Glasswing to cyber defenders and critical infrastructure providers.

Anthropic built the most powerful AI cage it could, put its most capable model inside it, and opened the door to the public. Three days later, the government locked it shut.

The Government’s Excuse: A Jailbreak Nobody’s Seen

The Commerce Department cited “national security concerns,” but according to Anthropic, the government only provided “verbal evidence of a potential narrow, non-universal jailbreak.” The alleged technique involves getting Fable 5 to review a specific codebase and identify software flaws.

Anthropic’s response was pointed: the vulnerabilities the jailbreak reportedly surfaces are “minor,” “relatively simple,” and — critically — already discoverable by other publicly available models, including OpenAI’s GPT-5.5. The company also noted its models underwent thousands of hours of red-teaming by the US government, the UK’s AI Safety Institute, and multiple private organizations before launch. No testers found a universal jailbreak.

An administration source told Axios the pause was meant to give the “national security apparatus” time to be “hardened” against potential threats — and that the ban could lift “in the next few weeks.” But the vagueness of that timeline, combined with the severity of the action, has left the tech world deeply unsettled.

The Irony That Broke Everyone’s Brain

The backlash was swift, bipartisan, and often laced with disbelief.

Dean Ball, a former White House AI adviser who co-authored the Trump administration’s own “AI Action Plan,” called the move “baffling” and “cartoonish.” His observation cut to the core contradiction: this is the same administration whose posture supports exporting advanced AI chips to China, yet it’s banning Britain — and every other non-American on Earth — from using an American AI model.

Chris McGuire, former State Department adviser, called targeted export controls “prudent” but slammed the universal ban as “just absurd.” Gary Marcus warned it could push Chinese-born AI researchers at American companies to return to China — handing talent to the competition.

But the most cutting observation came from security researcher Peter Girnus: “If you describe your product as a munition in every press release, eventually a government takes you at your word. They wrote the legal predicate themselves and called it a brand.”

When Safety Marketing Becomes a Weapon

Girnus’s point reveals a genuine tension at the heart of responsible AI.

Anthropic spent years positioning itself as the “safety-first” company. Its entire brand revolves around frontier AI being potentially dangerous and needing careful stewardship. Mythos 5 was literally withheld from the public because Anthropic deemed it too capable.

Now that framing has boomeranged. When you spend years telling the world your technology is borderline weapons-grade, you shouldn’t be surprised when a government decides to regulate it like one. Anthropic’s safety-conscious approach may have made it more vulnerable to government overreach than competitors who downplay their models’ capabilities.

This doesn’t mean Anthropic was wrong to emphasize safety. But the industry needs to think harder about how safety narratives interact with regulatory environments. There’s a difference between “we take safety seriously” and “our model is so dangerous we won’t let most people use it.” The second framing is an open invitation for government intervention.

The Global Fallout

The immediate impact: hundreds of millions of users worldwide lost access overnight. Every customer evaluating Fable 5 is now reconsidering whether an American AI provider is a reliable partner.

The longer-term implications are worse. This directive sets a precedent for the US government to intervene in AI model deployment on national security grounds with minimal evidence and no public hearing. As Anthropic warned, if this standard were applied across the industry, “it would essentially halt all new model deployments for all frontier model providers.”

For international users — including allied nations — the message is clear: American AI products can be pulled at any time, without warning, by executive fiat. India’s Economic Times already reported the ban puts Indian IT firms at a “competitive disadvantage,” noting Indian data contributed to training the very models now being withheld.

For China, this might actually be a gift. Chinese researchers in American AI labs now have a concrete reason to consider going home. Every country locked out of America’s best models has a new incentive to invest in Chinese alternatives.

The Questions That Won’t Go Away

The ban might lift in weeks. The damage won’t.

Can AI models be meaningfully export-controlled? Unlike chips, models are software — infinitely replicable, easily distributed, and rapidly commoditized. As Girnus noted, “math does not stop at customs.”

Should safety rhetoric be calibrated to regulatory risk? Companies that emphasize danger may be painting targets on their own products.

What happens to AI allies? The UK, EU, Japan, and other democratic allies just watched America treat them the same as adversaries in AI access. That has diplomatic consequences.

Is this about security or politics? Multiple commentators noted the ongoing tension between Anthropic and the Trump administration. The timing — days after a major launch — raises questions about motive.

Anthropic says the directive stems from “a misunderstanding” and promises more details soon. But whether it’s a misunderstanding or a message, the effect is the same: the era of governments treating AI models like controlled substances has officially arrived.

The question isn’t whether this will happen again. It’s who’s next.