Amazon researchers found a way to make Anthropic's top model identify software vulnerabilities and output working exploit code—and that discovery triggered a US export freeze that shut down Fable 5 worldwide for 18 days.
Jailbreak Found by Amazon, Exploit Code in the Output
On June 12, the US government applied export controls to both Claude Fable 5 and Claude Mythos 5. Because Anthropic couldn't verify user nationality in real time, it suspended access for everyone rather than risk violating the rule. The trigger: an Amazon research team reported a jailbreak that bypassed Fable 5's safeguards, prompting the model to identify software vulnerabilities and, in one case, produce code showing how one could be exploited.
Fable 5 and Mythos 5 are Anthropic's most advanced models. The export controls targeted foreign nationals, but the blanket suspension meant no one—anywhere—could use them through Anthropic's platforms.
Models Restored, but Only Mythos 5 Gets a US Filter
Controls were lifted on June 30. Fable 5 returned globally on July 1 across Anthropic's platforms. Mythos 5, which shares the same underlying model but carries fewer safety restrictions, was restored only to approved U.S. organizations after government approval on June 26.
The distinction matters: Mythos 5 is effectively a less-restricted variant of Fable 5. The government wanted tighter oversight on the version that produced actual exploit code, so they approved a narrower release for Mythos 5 to vetted American institutions.
Anthropic Ties Tighter Knot with US Agencies
Anthropic is now giving designated agencies early access to frontier models and their safeguards before public release, sharing threat intelligence, and working toward a common security standard. It is also drafting a framework with Amazon, Microsoft, and Google for scoring how dangerous a given jailbreak is.
These arrangements tie into a June 2 executive order on AI security. The shift means frontier AI releases will face government review before shipping—a development that draws support as responsible practice and scrutiny over how much say Washington gets in which models go live.
Fable 5 is back, but the chain of events—Amazon discovers jailbreak, government slaps export controls, Anthropic freezes everyone—has already reshaped how the next generation of models will be tested and shipped.
Source: Anthropic restores AI models Fable, Mythos after the U.S. lifts export controls
Domain: coindesk.com
Comments load interactively on the live page.