← Back to Model Beat
5Models·Jun 17

The White House Wants Anthropic to Block All Jailbreaks. That May Not Be Possible

The Trump administration has conditioned the potential rerelease of Anthropic's Fable 5 model on the company's ability to guarantee that its safety guardrails remain impenetrable. While officials demand a system that cannot be bypassed, security researchers caution that creating an unhackable AI model is technically unfeasible given the current state of adversarial testing. This standoff highlights a growing regulatory push to hold developers strictly accountable for model security, even as technical experts warn that perfect protection against jailbreaking remains elusive.

Covered by 3 sources

Related stories

ModelsFrom Chatbots to Collaborators: AI’s Next EraJun 15 · 39 sourcesModelsPredicting model behavior before release by simulating deploymentJun 16 · 3 sourcesModelsGoogle’s Gemini-Powered AI Home Speaker Goes on Sale June 25Jun 17 · 5 sourcesModelsZ.ai Launches GLM-5.2 With a Usable 1M-Token Context, Two Thinking-Effort Levels, and No Benchmarks at LaunchJun 14 · 10 sources