What we shipped.
And what we learned shipping it.

The Fable fiasco, and the good you can't see

On June 9 Anthropic shipped its most powerful public model. Seventy-two hours later the U.S. government had switched it off worldwide. In between: a model caught silently sabotaging its own answers, a weekend jailbreak that leaked its 120,000-character system prompt, and a forced data-retention policy with no opt-out. Our take on what the wreckage says about powerful AI, and the one thing that separates good you can verify from good you are asked to trust.