The world’s most capable public AI model was online for just three days before the U.S. government abruptly declared it a national security risk and forced it offline worldwide.

Launch: a record‑breaking model

On June 9, Anthropic released Fable 5, a public version of its Mythos 5 frontier model, marketing it as safer but still extraordinarily capable at cybersecurity tasks.¹ Within days it topped public leaderboards and “dominated every benchmark” against OpenAI’s GPT 5.5, becoming the strongest model available to ordinary users.²

Behind the scenes, however, Anthropic had already warned Washington that Mythos‑class systems could perform “superhuman feats of hacking,” and limited Mythos access to a vetted Project Glasswing consortium of governments and companies.³

Amazon raises alarms, the White House reacts

On June 12, Amazon security researchers told the White House they had prompted Fable 5 to produce information useful for cyberattacks, describing a “fix this code” workflow that surfaced known vulnerabilities.⁴ Axios later reported the company “called administration officials Thursday night to share a report showing how they were able to jailbreak and access portions of Anthropic's powerful new Mythos model that pose a national security threat.”⁵

That Friday afternoon, Commerce Secretary Howard Lutnick sent Anthropic an export‑control directive barring access to Fable 5 and Mythos 5 by “any foreign national … including foreign national Anthropic employees.”⁶ Anthropic says the letter “did not provide specific details of its national security concern” and was based on “verbal evidence of a potential narrow, non-universal jailbreak.”⁷

Global shutdown and competing narratives

To comply, Anthropic “abruptly disable[d] Fable 5 and Mythos 5 for all our customers,” while leaving its other models online.⁷ The company publicly pushed back: “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.”⁸

Administration officials, by contrast, framed the move as necessary export control: foreign governments, firms and individuals were blocked from Anthropic’s “most advanced AI models,” marking “an escalation in Washington's effort to treat cutting-edge AI systems as national security assets.”⁹

Some in Washington privately questioned Anthropic’s judgment, with one official saying “everybody said Anthropic was a bad actor … They screwed us,” while others suggested the model must “remain locked down until the U.S. government's national security apparatus is hardened.”¹⁰

Cybersecurity community backlash

Security professionals quickly warned the decision might backfire. An open letter signed by dozens of experts argued 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.”¹¹ Axios summarized their concern: in “trying to avert an AI hacking crisis, the Trump administration may end up making U.S. cyber defenses weaker.”¹²

Former Facebook CSO Alex Stamos and others noted that the exploit Amazon flagged—getting Fable to generate proof‑of‑concept code—“exists across other leading AI models, too,” including foreign systems, meaning attackers are unlikely to be similarly constrained.¹³

Anthropic’s stance: real risk, wrong tool

Anthropic acknowledges a dual‑use problem: “a great deal of advanced usage of AI models is dual use,” especially in cybersecurity and biology.¹⁴ But the company insists Fable’s safeguards are “substantially more effective than those of any previously deployed model” and that “no testers have yet been able to find a universal jailbreak.”⁷

In its own words, export controls applied this way could “essentially halt all new model deployments for all frontier model providers,” by treating narrow, non‑unique jailbreaks as grounds for recall.⁸

A governance stress test

The episode is widely seen as a test case for ad‑hoc AI governance. The Verge called it the first time U.S. export rules have been used “to control access to an AI model in this way,” warning that governing AI through “opaque, ad hoc interventions is unsustainable.”¹⁵

TechCrunch argued the crackdown “was never about an AI jailbreak” alone, pointing to “personality differences” and prior clashes over surveillance and weapons use.¹⁶ The Economist described Trump’s blocking of Anthropic as “capricious and chaotic,” emblematic of a volatile regulatory climate for frontier AI.³

As Anthropic executives shuttle between San Francisco and Washington seeking a truce,¹⁷ experts warn that similarly powerful—and jailbreakable—models are coming from competitors regardless.¹⁴ For now, the world’s top‑performing AI remains dark, sidelined less by its code than by a high‑stakes argument over who should decide when it is safe enough to use.