Trump Administration Demands Anthropic Block All AI Jailbreaks

The Trump administration has drawn a hard line with Anthropic: the AI company cannot rerelease its Claude Fable 5 model until it solves a problem that may have no solution.

According to details first reported by WIRED, administration officials told the publication that Anthropic must prevent users from circumventing the model's safety guardrails before Claude Fable 5 can return to service. The government took the model offline last week using export controls after the National Security Agency concluded that jailbreaking techniques could disable protections designed to prevent misuse of capabilities related to cybersecurity, chemistry, and biology.

Anthropic has maintained that the administration's concerns are overblown and that jailbreak effects are minimal. The company reiterated this position during a technical meeting Monday with the Commerce Department and the Office of the National Cyber Director, Sean Cairncross.

An Impossible Standard

But administration officials say the debate over jailbreak severity is over. Three people familiar with the discussions told WIRED that the government now views this as Anthropic's problem to fix. Neither the Commerce Department's Center for AI Standards and Innovation nor the National Security Agency has the resources to chase down every potential jailbreak on every model that reaches the market, these sources said.

The administration wants Anthropic to proactively and continuously test not just Fable 5 but all its frontier AI models for potential jailbreaks, then report vulnerabilities to the government.

The fundamental challenge: independent cybersecurity experts increasingly believe that unbreakable AI guardrails cannot exist. Skilled users and future AI models will inevitably find ways to bypass constraints, making guardrails only a temporary stopgap measure. What the White House appears to demand may be technically impossible.

Why it matters

This standoff reveals a critical gap between government expectations and technical reality in AI regulation. If the administration's standard becomes policy—requiring companies to guarantee their models cannot be jailbroken—it could effectively halt the release of advanced AI systems. No company can promise absolute security against determined adversaries or sophisticated prompt engineering. The dispute also highlights how the government is taking a more aggressive regulatory posture on AI safety, even as it lacks the technical capacity to evaluate every model itself. The outcome will likely set precedent for how other AI companies must demonstrate safety compliance before releasing powerful models.

The White House declined to comment on the matter, according to WIRED's reporting.