According to Matteo Wong in The Atlantic, the White House investigated a reported vulnerability in Anthropic's Fable model. Security expert Katie Moussouris reviewed the White House's findings and reached an unexpected conclusion: it wasn't actually a flaw. The issue involved prompt engineering—asking Fable to 'fix this code' instead of 'review for bugs'—to trigger different behavior. Moussouris determined Fable was working as intended, helpfully assisting with legitimate code security tasks.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton