The Trump administration has escalated its disagreement with Anthropic over the security of its most advanced AI model, demanding that the company prevent all jailbreaking vulnerabilities—a requirement that independent cybersecurity experts say may be technically impossible, according to a report by WIRED.
Government Pressure Mounts
Trump officials told the publication Inner Loop that if Anthropic wants to rerelease Claude Fable 5—taken offline last week with export controls over jailbreak concerns—the company must take steps to address alleged vulnerabilities, according to three people familiar with the discussions. Anthropic has maintained that the administration's concerns are overblown and that the effects of jailbreaks are minimal, a position it reiterated in a technical meeting on Monday with the Commerce Department and the Office of the National Cyber Director, Sean Cairncross.
NSA Confirms Vulnerabilities
The National Security Agency concluded that there are ways to disable guardrails on Fable 5, which are designed to prevent users from accessing capabilities of the underlying Mythos model related to cybersecurity, chemistry, and biology. Administration officials say they are now past arguing whether the jailbreaks are significant and view the situation as Anthropic's problem to fix.
Limited Government Bandwidth
Neither the Commerce Department's Center for AI Standards and Innovation nor the National Security Agency has the staff or bandwidth to chase down every conceivable jailbreak on every model that reaches the market, the sources said. The administration therefore expects Anthropic to be more proactive in continually testing not just Fable 5 but all its frontier AI models to find potential jailbreaks and flag them to the government.
Technical Obstacles Remain
Independent cybersecurity experts have increasingly taken the view that guardrails on AI models are only a stopgap solution, since skilled users and future AI models will find ways to bypass constraints. This suggests that what the White House appears to want—complete prevention of jailbreaking—may not be possible. A White House spokesperson declined to comment.
The standoff underscores the challenges regulators face as advanced AI models proliferate, and it raises questions about the feasibility of government mandates for AI safety in the absence of clear technical solutions.
Sources:
