According to security researcher Kate Moussouris, cited by tech journalist Simon Willison, the US government has banned Fable 5 under export controls based on a misunderstanding of how AI models help with cyber defense. Researchers asked the model to review code containing known vulnerabilities—a standard security practice—and then to fix them. The model refused the first request but succeeded at the second, which prompted regulators to restrict it as a national security threat. Moussouris argues this logic is backwards: asking AI to find and fix security bugs is precisely what defenders do every day. She writes, 'That is not a guardrail bypass. It is the most valuable thing an AI model can do for defensive security.' The irony: in attempting to prevent AI-enabled attacks, policymakers may have actually blocked a critical tool for defending against them.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton