US Government Orders Anthropic to Restrict Access to Advanced AI Models The abrupt U.S. shutdown of Anthropic’s most powerful AI models has turned a technical dispute into a global reckoning over who controls frontier AI.

On June 9, Anthropic released Claude Fable 5, a public, heavily safeguarded version of its Mythos‑class cybersecurity model, alongside the restricted Mythos 5 for vetted defenders. The company had warned Mythos was “too powerful” for broad release because of its superhuman ability to find software flaws, while pitching Fable as “safe for general use.”

Within days, Amazon security researchers said they could rephrase prompts so Fable would “fix this code” and surface known vulnerabilities, a behavior critics argued exists in other top models as well. On June 12, after frantic calls from Amazon CEO Andy Jassy to senior U.S. officials, the Commerce Department invoked an obscure export control, ordering Anthropic to block Mythos 5 and Fable 5 for “any foreign national,” including its own staff. With no practical way to separate users, Anthropic pulled both models worldwide within about 90 minutes of being notified.

In Washington, officials cast the move as a national‑security necessity amid fears that Fable’s guardrails could be bypassed and that China‑linked actors might copy Mythos’ capabilities. The episode is the first time a government has forced withdrawal of a publicly released frontier model, and commentators describe it as an assertion of American gatekeeper power over “the world’s most important technology.”

Anthropic disputes that the issue amounted to a true jailbreak, calling it “a narrow, already‑patched” flaw that revealed only minor, publicly known bugs. Cybersecurity veterans back that view: an open letter signed by roughly 100 experts argues the action “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.” Former Facebook security chief Alex Stamos warns the precedent tells U.S. labs “American models can’t do defensive security research.”

As the weekend crisis deepened, Anthropic flew senior security staff to D.C. for talks with Commerce and the White House, while a separate Axios account quoted an administration official saying “everybody said Anthropic was a bad actor… They screwed us,” underscoring personal frictions as much as technical ones. Business Insider later reported that the dispute has shifted into negotiations over a shared framework to grade the severity of model “jailbreaks” and guide future interventions.

Abroad, the cutoff landed like a warning shot. India, Anthropic’s second‑largest market, saw revenue growth halted overnight and a major Tata Consultancy Services deal thrown into doubt, turbo‑charging calls for a multibillion‑dollar sovereign‑AI fund and a pivot to open‑source models. European startup Mistral framed the saga as proof that foreign providers – and their governments – “have the keys” to critical AI, bolstering its pitch for locally deployable, open‑weight systems. Commentators in Europe and Asia argue the U.S. move has “handed” rivals like Mistral and Chinese lab DeepSeek a strategic opening while gifting Beijing a stronger case for its own models.

All sides converge on one point: ad‑hoc, opaque export directives are an unstable way to govern rapidly advancing AI. Analysts say the Anthropic clash has exposed AI’s “missing referee,” turning abstract talk of oversight into a real‑time test of how far governments will go – and how the rest of the world will respond.

Continue reading https://foxvector.com/stories/019ee6b0-223a-0a3b-70a9-37b44f9eb5e5