US Orders Anthropic to Shut Down Fable 5 and Mythos 5 Worldwide

Anthropic disabled its two most capable AI models — Claude Fable 5 and Claude Mythos 5 — for every customer worldwide on Friday evening, after the US government issued an export control directive barring access by any foreign national. The order, sent by Commerce Secretary Howard Lutnick to Anthropic CEO Dario Amodei at 5:21 PM ET, came just three days after the models launched, according to Tom's Hardware.

The directive specifically targets Mythos‑class models and requires a license for the export, re‑export, or domestic transfer of the technology. Crucially, it extends the restrictions to any foreign person on US soil — including Anthropic's own employees. With no reliable way to screen foreign nationals out of its user base in real time, Anthropic pulled both models globally rather than attempt a partial block.

The sweeping government action was reportedly triggered by a direct intervention from Amazon CEO Andy Jassy. According to Reuters, Jassy was among tech leaders who raised concerns to senior Trump administration officials about security risks in Anthropic's latest models. The Wall Street Journal reported that the information Jassy shared sparked the abrupt, sweeping move to halt foreign access.

Amazon is both an Anthropic investor — having poured billions into the company — and a competitor through its own AI efforts. The irony of an investor triggering a government crackdown on its own portfolio company is not lost on industry observers.

What exactly prompted the national security concern? Anthropic said the government supplied only verbal evidence pointing to what Fortune characterized as a "narrow, non‑universal jailbreak." The technique reportedly involved asking the model specific prompts that could bypass safety guardrails — but the research was not designed as a jailbreak and did not give the model dangerous new capabilities.

One cybersecurity CEO told Fortune the research "wasn't geared towards jailbreaking" and that the Commerce Department's response appears disproportionate to the actual risk. Anthropic, for its part, has not been given a detailed written explanation of the evidence.

The shutdown is the culmination of months of escalating friction between Anthropic and the Trump administration. In February, the White House barred all federal agencies from using Anthropic products. That same day, Trump called Anthropic "a radical left, woke company" amid a dispute over the company's refusal to permit the military to use its technology for what CEO Dario Amodei described as mass surveillance of Americans and fully autonomous weapons, per.³

Despite those tensions, Anthropic filed for an IPO earlier this month. The timing of this export control order — days after the IPO filing and three days after the company's most significant model launch — raises questions about whether the government is using regulatory tools to pressure a company it views as politically uncooperative.

For the thousands of developers and companies that integrated Claude Fable 5 into their products over those three days, the shutdown is a stark reminder of AI supply chain risk. API calls to Fable 5 and Mythos 5 now fail. Any application depending on those models is broken until developers migrate to older, less capable Claude models — or switch providers entirely.

The order's broad language — covering any foreign national "whether inside or outside the United States" — means even US‑based startups with international team members cannot legally access the models. Anthropic employees who are foreign nationals are also locked out of their own technology. As Forbes noted, the lockdown exposes growing enterprise AI risk: companies that build on most advanced models may find those models removed overnight with no warning and no recourse.

Anthropic said in its statement that it is working with the government to resolve the restrictions, but no timeline has been given. The company faces a difficult choice: comply with a government it has publicly clashed with, or fight the order and risk further escalation that could jeopardize its upcoming IPO.

For the broader AI industry, the shutdown sets a precedent. If the US government can disable commercial AI models worldwide with a single directive — based on verbal evidence of a narrow vulnerability — then every AI company's product roadmap exists at the pleasure of the Commerce Department. TIME described the move as one that "could reshape the global AI landscape" by forcing companies to build compliance architectures that segment users by nationality — a technical and ethical thicket that no company has solved.

In the meantime, developers are left with older Claude models, or looking across the fence at competitors like GPT‑5.5, which Business Insider noted now faces less competition at the high end of the market.