White House Demands Anthropic Fix AI Jailbreaks Experts Say Are Unfixable

Executive Briefing

• Requires Anthropic to prevent all jailbreaks on Claude Fable 5 before allowing its rereleased after export control removal.
• NSA concluded guardrails on Fable 5 can be disabled, exposing cybersecurity, chemistry, and biology capabilities.
• Independent experts warn AI guardrails are only stopgap measures, meaning the White House's demand may be technically impossible.
• Administration shifts responsibility entirely to Anthropic, expecting proactive jailbreak testing and government reporting on all frontier models.
• Commerce Department and NSA lack staff to chase jailbreaks across every model reaching the market, officials say.