Anthropic Pulls Back Claude Fable 5 And Mythos 5 Models As US Orders Takedown Over Security Concerns

Hyderabad: Anthropic earlier this week launched two of its most powerful Artificial Intelligence (AI) models, Claude Fable 5 and Claude Mythos 5, for all users worldwide. However, the US government on Friday ordered the Amodei-led AI company to immediately shut off worldwide public access to these AI models, highlighting national security concerns.

The AI company received the directive at 2:51 AM IST on Saturday, June 13, 2026, which translates to 5:21 PM ET on Friday, June 12. It instructed Anthropic to suspend all access to the models "by any foreign national, whether inside or outside the US, including foreign national Anthropic employees."

Anthropic announced on X that it has complied with the order, but made clear it believes the government's reasoning is flawed. It highlighted that "Access to all other Claude models is not affected."

Why these AI models matter

Both models originate from Claude Mythos Preview, a highly advanced model intended for security research, which was capable of finding security bugs and flaws. Access to Mythos Preview was initially limited to a small group of companies and research partners through Project Glasswing. When Anthropic unveiled Claude Mythos 5 and Claude Fable 5, the company said the Mythos family would remain restricted to Project Glasswing participants, while Fable would be made publicly available.

Anthropic launched Fable 5 with safeguards that route sensitive queries to Claude Opus 4.8 instead, with topics tied to cybersecurity, biology, chemistry, and distillation able to trigger that fallback, which Anthropic says fires in less than five per cent of sessions.

Anthropic launched Fable 5 with safety measures in place. If a user asked query related to cybersecurity, biology, chemistry, or distillation, the model passes it to Claude Opus 4.8. Anthropic says this fallback occurs in fewer than five per cent of sessions.

The Jailbreak Claim

Anthropic's understanding is that the government believes it has become aware of a method of bypassing, or "jailbreaking", Fable 5. However, the company disputes the severity of the finding. Anthropic describes the demonstrated technique as "a narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws," and says it surfaces previously known, minor vulnerabilities that are discoverable by other public models — including GPT-5.5 — and used daily by legitimate security professionals.

Ahead of launch, the company red-teamed Fable's safeguards for thousands of hours with the US government, the UK AISI, third-party organisations, and internal teams, and says testers have not found a universal jailbreak.