The United States government has ordered AI safety startup Anthropic to immediately suspend access to its most advanced artificial intelligence models, Claude Fable 5 and Claude Mythos 5, citing national security concerns. The export control directive, which targets all foreign nationals both inside and outside the U.S., has forced Anthropic to disable the models globally.
The suspension directive was delivered to Anthropic on June 12, 2026, at 5:21 PM Eastern Time (ET), less than a week after the models were officially released on June 9. According to Anthropic, the government’s order did not contain detailed specifications regarding the exact national security threats posed by the models. Because Anthropic cannot technically verify the nationality of its users in real-time, the company was forced to disable access for all users worldwide, including those in international markets.
The Jailbreak Dispute
The U.S. government’s action is reportedly based on claims that researchers discovered a bypass or “jailbreak” method for the Fable 5 model. However, Anthropic stated that the only evidence provided by authorities was verbal and described a narrow, non-universal jailbreak technique. This specific exploit involved asking the model to analyze a particular codebase and repair software vulnerabilities within it.
Anthropic has contested the severity of the government’s claims. Following an internal review of the reports, the company concluded that the capabilities demonstrated in the jailbreak are already publicly available in other frontier models, such as OpenAI’s GPT-5.5.
Advanced Capabilities and Safeguards
Claude Fable 5 was the first public model released under Anthropic’s new Mythos class, which sits above the Opus class in the company’s capability hierarchy. Prior to the public release, a preview version of Mythos demonstrated the ability to detect thousands of critical security vulnerabilities, bugs, and exploits across major operating systems and web browsers.
To mitigate these risks, Anthropic designed Fable 5 with strict guardrails covering high-risk domains, including cybersecurity, biology, chemistry, and nuclear distillation. When encountering restricted queries, the model was programmed to block the response and redirect the session to Claude Opus 4.8.
Anthropic expressed strong disagreement with the global suspension, arguing that a narrow, non-universal exploit should not justify pulling a commercial model used by hundreds of millions of people. The company warned that if such a standard were applied across the industry, it would effectively halt the release of any new frontier AI models.
Despite the disagreement, Anthropic is complying with the legal order. The suspension only affects Fable 5 and Mythos 5, which are priced at 10 USD per million input tokens and 50 USD per million output tokens. Access to other models, including Claude Opus 4.8, Claude Sonnet 4.6, and Claude Haiku 4.5, remains unaffected. Anthropic stated that it is actively working with government officials to resolve the misunderstanding and restore access as soon as possible.