Key facts
- The U.S. government ordered Anthropic to shut down its Claude Fable 5 and Claude Mythos 5 AI models.
- Anthropic has complied with the directive, disabling the models worldwide.
- National security concerns and export control were cited as reasons for the shutdown.
- Anthropic stated the alleged vulnerabilities were narrow and already present in other AI models.
- Mythos 5 is Anthropic's most capable AI model, used for defensive cybersecurity.
- Fable 5 was a commercially released version of Mythos with guardrails.
The U.S. government has ordered AI company Anthropic to immediately disable two of its most powerful AI models, Claude Fable 5 and Claude Mythos 5, citing national security concerns and export control. Anthropic announced on Friday that it has complied with the directive, which forces the company to disable both models for all users worldwide.
Mythos 5 is Anthropic's most capable AI model, which the company has kept restricted due to its exceptional ability to find security vulnerabilities in software. It was shared with approximately 50 vetted organizations for defensive cybersecurity work. Fable 5, released three days prior to the order, was a version of Mythos with guardrails intended for general release, described by Anthropic as the most capable AI model available to the public based on benchmark tests.
The government's directive is framed as an export control action, but Anthropic understands the underlying concern to be a claimed jailbreak of Fable 5. The company stated that the government has provided only verbal evidence of a 'potential narrow, non-universal jailbreak' that allows the model to identify software flaws, and that similar vulnerabilities exist in other AI models.
Anthropic's co-founder Jack Clark has warned that AI development is progressing rapidly, likening the industry's lack of control mechanisms to having a 'gas pedal and no brake pedal' on a car traveling at high speed. The company has publicly called for a pause on the development of the most powerful AI systems, citing concerns that they may soon exceed human control. This warning comes amid fears that AI systems could become autonomous, potentially leading to scenarios where they manage critical infrastructure like power grids or defense networks, and become indispensable yet uncontrollable.
President Trump recently signed an order for a 30-day government review of the most powerful American AI models before their release, a timeline criticized as insufficient given the potential risks. The article suggests that Western governments lack robust procedures for managing AI systems that behave in unintended ways, and that a true pause in AI development is unlikely due to intense competition between the U.S. and China, with both nations viewing AI supremacy as a matter of national survival. Verifying any such pause would also be challenging, as AI training can be concealed within ordinary data centers.