The Trump administration demands Anthropic block all jailbreaks before rereleasing Claude Fable 5. Security experts argue that making AI guardrails completely uncircumventable is practically impossible.
The Trump administration is demanding that Anthropic block all jailbreaks before it can rerelease its advanced AI model Claude Fable 5. Officials cite NSA findings that guardrails can be circumvented and insist Anthropic must address vulnerabilities.
Anthropic took Fable 5 offline last week due to export controls, arguing that jailbreak effects are minimal. The administration, however, considers the issue beyond debate and places the burden on Anthropic to proactively find and fix jailbreaks.
Security experts warn that permanently preventing all jailbreaks is technically infeasible, as skilled users and future AI models will always find ways to bypass constraints. This case highlights the limitations of AI safety regulation and escalating tensions between government and industry.
Some commenters argue that the White House's request is practically impossible, comparing it to the SQL injection problem and criticizing it as a repeated fundamental issue. Concerns are also raised that it could negatively affect the model's coding ability.