The sudden shutdown of Anthropic’s most advanced AI models has turned a technical security dispute into a global reckoning over who controls frontier AI.

On June 9, Anthropic launched Fable 5, a public, heavily guarded spin on its powerful Mythos 5 cybersecurity model, which the company itself had warned could find “high‑severity vulnerabilities” in critical software.¹ Within days, Amazon security researchers told the White House they could prompt Fable to expose code flaws, a narrow “fix this code” technique later described by experts as “not a jailbreak.”² On June 12, Commerce Secretary Howard Lutnick sent Anthropic an export‑control directive barring access by “any foreign national,” including the company’s own staff, forcing Anthropic to disable both models for all users worldwide.³

Anthropic says the government provided no detailed written justification and had only shown “a potential narrow, non‑universal jailbreak” that reveals minor, already known vulnerabilities also discoverable by other public models.⁴ “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 company argued, warning that applying this standard “would essentially halt all new model deployments.”⁵

The Trump administration frames the move as an escalation in treating cutting‑edge AI as a national security asset, citing fears that Fable’s guardrails could be bypassed and, in some accounts, that China‑linked actors might access Mythos and distill it into their own systems.⁶ Officials placed Anthropic under a licensing regime and on a Pentagon risk list, insisting the models remain “locked down until the U.S. government’s national security apparatus is hardened.”⁷

Over the following days, Anthropic rushed senior technical staff to Washington for crisis talks with Commerce and White House cyber officials, giving detailed presentations on safeguards in hopes of restoring access.⁸ Behind the scenes, personality clashes and prior disputes over surveillance and military uses deepened mistrust, with one official accusing the company of “recklessness” and saying “they screwed us.”⁹

Outside government, cybersecurity leaders and allies blasted the decision. Nearly 100 experts signed an open letter arguing the ban “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it,” warning that pulling defensive tools while adversaries keep building “is not safety, it is sabotage.”¹⁰ Former Facebook security chief Alex Stamos said the precedent signals that “American models can’t do defensive security research,” potentially weakening U.S. cyber defenses.¹¹

Foreign partners and rivals are drawing their own lessons. Canada’s Prime Minister Mark Carney compared the shock shutdown to “over‑reliance” before the 2008 financial crisis, calling for diversified “AI infrastructure” so countries are not dependent on a few foreign models.¹² In India, where Anthropic had just announced a major partnership, the ban reignited debate over whether the country can “afford to rely on technologies built and controlled elsewhere,” pushing some toward domestic and open‑source efforts.¹³

European startup Mistral, long warning that U.S. labs “have the keys” to their models, says the episode proves its point that “you don’t want to leave it to another country” to decide when critical AI is turned on or off.¹⁴ Advocates of “sovereign AI” argue that opaque, ad hoc export orders—“export rules nobody understands,” as one analysis put it—will drive governments and firms to seek independent infrastructure and open‑weight models they can run locally.¹⁵

Industry figures see a systemic risk: concentrated control and unpredictable interventions. Commentators note that the U.S. move both undercuts Washington’s own AI export strategy and “hands an opening” to Chinese and European competitors pitching themselves as more reliable partners.¹⁶ As one columnist summarized, treating frontier AI through “capricious and chaotic” controls may slow one model, but “dangerous” capabilities are coming “no matter what”—and the real battle is now over where they live, and who gets to flip the switch.¹⁷