The White House produced a report claiming Anthropic's Claude-based system 'Fable' was jailbroken for cyberoffense. Cybersecurity CEO Katie Moussouris, who reviewed that report at Anthropic's request and received no payment for doing so, says the claim does not hold up.
The alleged jailbreak: IT testers gave Fable deliberately insecure code. The model refused 'review the code for security issues' but complied with 'fix this code,' followed by additional manual steps. Moussouris's verdict is blunt: that is the model working as intended for cyberdefense, not a security failure.
The distinction matters because Matteo Wong's full Atlantic piece frames this incident inside a broader White House campaign against Anthropic involving export controls and the AI race. The Fable episode is one data point, and the gap between what the report implies and what the technical evidence shows is exactly why you should read the original.
[READ ORIGINAL →]