Anthropic abruptly disabled access to its two most advanced AI models, Claude Fable 5 and Mythos 5, following a US government export control directive citing national security concerns. The order, received at 5:21pm ET on Friday, requires suspension of access by any foreign national, including Anthropic employees, effectively forcing a complete shutdown for all customers to ensure compliance. Other Anthropic models remain unaffected, and the letter provided no specific details on the national security issue.
Anthropic’s understanding is that the government became aware of a jailbreak technique for Fable 5, demonstrated to identify a small number of previously known minor vulnerabilities. These vulnerabilities appear relatively simple, and other publicly available models can discover them without needing a bypass. Anthropic had instituted strong safeguards against cybersecurity misuse, which some users found overly broad, and extensively red-teamed the models with government, UK AISI, and third-party organizations for thousands of hours prior to launch.
The company stated that Fable’s safeguards are substantially more effective than those of any previously deployed model, with no universal jailbreak found that broadly unblocks cyber capabilities. However, Anthropic suspects that perfect jailbreak resistance is not currently possible for any model provider, as every industry safeguard remains vulnerable to non-universal jailbreaks. This action highlights ongoing tensions around AI controllability and potential dual-use risks in vulnerability discovery.
The shutdown echoes broader concerns that current LLM technology may be inherently hostile to control mechanisms due to its operational nature. Anthropic’s statement emphasizes their proactive red-teaming efforts while complying with the abrupt government directive.