US halts latest Anthropic AI models over security concerns

The US government has ordered Anthropic to disable access to its two most advanced AI models—Claude Fable 5 and Claude Mythos 5—just days after their public release. The shutdown, effective June 12, 2026, follows an export control directive citing national security concerns. The directive bars access for any foreign national, including Anthropic’s own employees, prompting the company to suspend the models entirely rather than risk compliance violations.

A narrow vulnerability with broad consequences

Anthropic’s statement suggests the government’s action stems from a demonstration of a "jailbreak" technique—bypassing safeguards to elicit restricted behavior—observed in Fable 5. The company argues this flaw is minor, widely documented in other models like OpenAI’s GPT-5.5, and already addressed through existing safeguards. Anthropic contends the government’s response misinterprets the risk, warning that such a broad recall standard could stall AI innovation across the industry.

The models caught in the middle

Fable 5 and Mythos 5, both part of Anthropic’s premium Mythos-class, were designed with layered safeguards, including classifiers that detect misuse in areas like cybersecurity and biology. Fallback mechanisms redirect flagged queries to the safer Claude Opus 4.8 in under 5% of sessions. Despite these protections, the government’s order halts their availability entirely, leaving only restricted access for trusted partners under the Glasswing program.

While other Anthropic models remain unaffected, the episode highlights the tension between rapid AI deployment and regulatory oversight. Anthropic is complying with the directive while publicly challenging its rationale, framing the situation as a potential misunderstanding that could set a precedent for future model launches.

---

Source: MarkTechPost. AI-assisted editorial synthesis — TechnoExpress.