Anthropic has been ordered by the US government to suspend all access to its most capable cybersecurity models, Fable 5 and Mythos 5, citing national security concerns under export control authorities.
The directive, received at 5:21pm ET on June 12, requires Anthropic to disable access for all foreign nationals, whether inside or outside the United States, including foreign national Anthropic employees. In practice, Anthropic says it had no choice but to pull the models for all customers entirely.
Access to all other Anthropic models, including Claude, remains unaffected.
What triggered this? The government indicated it had become aware of a jailbreaking technique capable of bypassing Fable 5’s safeguards. Anthropic says the demonstrated technique identified a small number of previously known, minor vulnerabilities, and that other publicly available models can discover them too without needing a bypass.
Anthropic pushed back on the characterization. The company noted that Fable 5’s safeguards were red-teamed for thousands of hours alongside the US government, UK AISI, and multiple third-party organizations before launch. No tester, they say, ever found a universal jailbreak. “Perfect jailbreak resistance is not currently possible for any model provider,” Anthropic stated, adding that every industry safeguard is vulnerable to targeted, non-universal jailbreaks.
The company had adopted a defense-in-depth strategy instead: making jailbreaks narrow or expensive to produce, combined with monitoring to detect and shut down successful attacks quickly.
This is an unprecedented move. No AI company has had a model pulled by government export order before. It sets a remarkable precedent for how the US government treats powerful AI models under existing arms export frameworks, and it raises serious questions about what happens when commercial AI capabilities intersect with national security claims.
