The U.S. government ordered Anthropic on Friday to immediately revoke access to two of its most advanced AI systems — Claude Fable 5 and Claude Mythos 5 — over national security worries. Anthropic stated on X that it has complied with the order, while making clear that it believes the government erred in this case. The directive — which the company said it received Friday at 5:21 p.m. ET — requires Anthropic to disable both models for all users globally, not merely the foreign nationals targeted by the export control order. Access to Anthropic’s other models remains unaffected. So why does any of this matter? Anthropic’s Mythos is its most advanced AI model to date. The company first revealed it in early April but has since kept it under strict access controls, citing the model’s exceptional ability to discover security vulnerabilities in software. According to Anthropic, Mythos uncovered vulnerabilities in every major operating system and web browser it was tested on. Instead of releasing it publicly, the company initiated a limited program called Project Glasswing, distributing the model to approximately 50 carefully selected organizations—including Amazon, Apple, Google, Microsoft, and CrowdStrike—for use in defensive cybersecurity efforts. Fable 5, which launched just three days ago, represents Anthropic’s response to mounting commercial pressure: a version of Mythos equipped with guardrails that prevent it from answering questions in sensitive domains such as cybersecurity and biology, which the company says makes it safe for widespread release. It instantly became the top-performing AI model accessible to the public, based on benchmark evaluations conducted by Vals AI, a firm that monitors AI technology performance. Image Credits: Vals AI. The government’s measure is presented as an export control measure that limits foreign nationals’ access to the models. In a detailed blog post, Anthropic states that it believes the core issue stems from an alleged jailbreak of Fable 5.
Anthropic’s safety warnings may have just backfired — the government has pulled the plug on its most powerful AI
- by stefan