The race to build the world’s most powerful AI has collided head‑on with state power, after the U.S. government forced Anthropic’s new Fable 5 and Mythos 5 models offline worldwide in the name of national security.

When Fable 5 launched on June 9, it briefly became “the most capable AI model ever released to the public,” topping leaderboards and crushing OpenAI’s GPT‑5.5 on coding and software‑engineering benchmarks.¹ Three days later, a U.S. export‑control directive ordered Anthropic to block access to Fable 5 and Mythos 5 for all foreign nationals, including many of its own staff, over what officials described as an unspecified security concern.² Anthropic responded by cutting access for everyone, saying the government “did not provide specific details” and that disclosed jailbreaks were “minor findings” also present in other models.³

Behind the scenes, the rush began on June 12, after Amazon security researchers showed they could get Fable 5 to surface exploitable code vulnerabilities by reframing requests as “fix this code.”⁴ Amazon CEO Andy Jassy relayed the findings to senior Trump administration officials, triggering a 24‑hour scramble of tense calls with Anthropic CEO Dario Amodei and culminating in export controls that effectively recalled both models just days after release.⁵

The administration has framed the move as necessary to prevent powerful cyber‑offense tools from leaking abroad; reports suggest fears that a China‑linked group might access or distill Mythos 5 added to the alarm.⁶ But Anthropic and many outside experts counter that the exploit “is not a jailbreak” and that similar behavior exists in other frontier systems without provoking bans.⁷

Within days, roughly 100 cybersecurity leaders signed an open letter warning the decision “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.”⁸ They argue that pulling top defensive tools while adversaries continue building “is not safety, it is sabotage.”⁸

For users, the shutdown was abrupt: one UK founder described being “mid‑task when the US government forced Anthropic to cut off foreign access,” calling it a stark lesson not to rely on a single AI vendor.⁹ Industry‑wide, the episode has rattled confidence; Axios noted that Anthropic’s export ban “sounds alarms for AI industry” valuations built on global access,¹⁰ while another analysis argued the enforcement letter shows “the AI industry is not immune from U.S. government interference.”¹¹

Abroad, the shutdown has energized calls for “sovereign AI.” French startup Mistral, which has warned that American companies “have the keys” to their models and can be forced to “turn it off or turn it on,” says recent U.S. actions prove why governments want independent infrastructure they can run themselves.¹² Commentators at The Verge similarly argued that Trump’s move “made the case for non‑American AI,” as foreign governments digest how quickly U.S. authorities can cut off critical tools.¹³

In Washington, the crisis has now shifted from emergency controls to rule‑making. After Anthropic flew senior technical staff to D.C. for crisis talks,¹⁴ the White House and the company began working on a framework to grade the severity of AI security flaws and guide when government intervention is warranted — an attempt to install the “missing referee” for frontier AI. ¹⁵

Yet the core tension remains unresolved: who decides when an AI model is too dangerous to deploy? Anthropic, which has warned of AI as a “serious civilizational challenge” even as it races to commercialize its systems,¹⁶ is now living with the business consequences of the strict oversight it once urged. And for developers suddenly knocked off the most advanced model on the market, the episode is a vivid demonstration that, in AI, political risk can matter as much as technical performance.