The Trump Administration shut down Anthropic's Claude Fable 5 and Mythos 5 models on Friday, citing what Amazon reported as a 'jailbreak' discovered in Fable. The supposed exploit: asking the model to review code for security vulnerabilities—which it did, identifying flaws that other advanced AI systems also found easily. According to Zvi at LessWrong, the Administration now demands Anthropic 'fix' this flaw, but the author argues this is logically impossible. A model either possesses the skill to write secure code and identify its weaknesses, or it doesn't. You cannot selectively ban the defensive use of that capability. The shutdown is now in its seventh day, with resolution prospects uncertain—the author estimates roughly even odds the pause will lift by July 1st.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton