US Orders Anthropic to Disable AI Models Over Jailbreak Concerns

Anthropic will immediately shut down access to its most advanced AI models following a US government directive that cites national security concerns over a potential exploit.

The Commerce Department issued an export control order requiring Anthropic to suspend access to its Fable 5 and Mythos 5 models for all foreign nationals. To ensure compliance, Anthropic said it must disable the models for all users globally. Other Anthropic models remain unaffected.

According to Anthropic, the government believes a method exists to bypass safeguards that prevent Fable 5 from identifying software vulnerabilities in code. The company said it received only "verbal evidence of a potential narrow, non-universal jailbreak" and was not provided specific details about the national security concern.

Why it matters

This marks the first time US export controls have targeted deployed AI models themselves rather than the chips and infrastructure that power them. The action signals a significant shift in how regulators approach AI risk assessment and establishes precedent for recalling commercial AI products based on potential exploits. For companies preparing to go public—Anthropic filed confidentially for an IPO last month—the incident demonstrates how national security considerations can override commercial deployment decisions.

Escalating tensions with government

The order arrives amid deteriorating relations between Anthropic and US authorities. Earlier this year, Anthropic refused to allow the military to use its models for domestic surveillance and fully autonomous weapons systems. In response, the government placed Anthropic on a supply chain blacklist scheduled to take effect later in 2026.

Anthropic released Claude Fable 5 earlier this week as its first "Mythos-class" model, representing a new capability tier. The company implemented guardrails restricting use in areas including cybersecurity, though some users found these protections overly restrictive.

Experts have warned that Mythos-class models could accelerate sophisticated cyberattacks, particularly against sectors like banking that depend on complex legacy technology systems.

Company disputes government assessment

Anthropic contested the government's decision, arguing that discovering a narrow potential jailbreak should not justify recalling a model deployed to hundreds of millions of users. The company noted it worked with the US government on safety protocols before launching Fable 5 and that competing AI models demonstrate similar abilities to identify minor code bugs.

As recently as the day before the order, Anthropic had advocated for stronger US oversight of AI, including authority to block models with unacceptable risks. However, the company said Friday's action failed to follow principles of fair, fact-based regulation.

Pentagon Chief Information Officer Kirsten Davies defended the decision on social media, stating that national security takes precedence over "revenue cycles, clickbait and pre-IPO valuation."

Anthropic characterized the situation as a "misunderstanding" and said it is working to restore model access. A US official confirmed the Commerce Department issued the export control directive.

These details were first reported by The Guardian.