The Once And Future Fable #3: Fix This Code

TL;DR

• 1

  There was no real jailbreak: Katie Moussouris, reportedly the only outside expert to read the paper, says the key prompt was simply "fix this code," applied to vulnerable code with several manual steps afterward.

• 2

  Fable gave no special uplift over rivals in the reported setup: The transcript claims Claude Fable 5 did not outperform Opus 4.8 or GPT 5.5 in the engineered scenario, which undercuts the idea that a unique dangerous capability escaped containment.

• 3

  The export control was meant to force a shutdown worldwide: The Bloomberg letter text reportedly required individually validated licenses for Mythos and Fable exports, reexports, and transfers to any destination or foreign person, and the WSJ quotes Lutnick's response as: "That's the point."

• 4

  Amazon's role looks muddled and possibly mischaracterized: Financial Times says Andy Jassy raised broad frontier AI concerns, while other accounts suggest Amazon tested Fable at White House request and a game-of-telephone effect turned harmless findings into a crisis.

• 5

  The UK and even Anthropic's own foreign staff got caught in the blast radius: Zima highlights that the controls reportedly denied a carveout for British companies and nationals, and even complicated internal Anthropic access, which he treats as evidence the policy was wildly overbroad.

• 6

  The bigger consequence is a de facto permission regime for frontier models: Even if Fable returns, the episode signals to every US lab that releasing a top-tier model without first asking the government may now be too risky.