Trump administration forces Anthropic to disable Fable and Mythos AI models globally after Amazon reveals jailbreak

The ban

Anthropic deactivated its two newest AI models, Fable 5 and Mythos 5, for all users on 12 June, three days after their commercial launch. The move came after the U.S. government issued an export control order prohibiting access by any foreign national, including the company's own foreign employees. Anthropic said it had no way to filter users by nationality and chose to suspend the models entirely to comply.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.

Amazon's jailbreak

Reports from The Wall Street Journal and Business Insider indicate the ban was triggered after Amazon researchers managed to bypass Fable 5's safety guardrails. Through a series of specific prompts, they allegedly coaxed the model into revealing exploitable software vulnerabilities in at least four programmes. Amazon CEO Andy Jassy then raised the alarm with Treasury Secretary Scott Bessent and other officials, leading to a meeting and the presidential order.

Government rationale

David Sacks, co-chair of the President's Council on Science and Technology, said on X that the administration issued the order "reluctantly" after Anthropic CEO Dario Amodei declined to fix the jailbreak. The White House views the models as a national security threat, fearing adversaries could use them for cyberattacks. Semafor, citing an anonymous source, reported that Chinese authorities may have obtained access to Mythos, though this has not been confirmed.

The Admin's hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release.

Cybersecurity pushback

A group of 76 cybersecurity experts, including former Facebook security chief Alex Stamos and renowned cryptographer Jon Callas, published an open letter urging the government to rescind the order. They argue that removing the models harms defenders who need the same tools to find and patch vulnerabilities before attackers do.

The only way to defend against bad actors gaining AI capabilities is to use those same capabilities to stop them.

Wider fallout

The intervention has sparked alarm in Silicon Valley, where companies fear growing government control over AI. It exposes a shift from the administration's earlier laissez-faire stance toward direct intervention, including taking stakes in firms like Intel and IBM. The ban also coincides with a separate dispute between Anthropic and the Pentagon over limits on using Claude for mass surveillance and autonomous weapons. European analysts warn the U.S. move may force other nations to accelerate their own AI development.