The most powerful AI in the world lasted thirteen days on the market.
In June 2026, Anthropic launched Claude Fable 5 - a model so capable it could rebuild software from a screenshot and compress two months of engineering into a single day. Then the U.S. government switched it off. Not throttled, not fined: switched off, worldwide, for every user on the planet, on the strength of a report the company's own largest investor delivered straight to the Treasury Department.
Switched Off reconstructs that month and everything it laid bare. Here is the jailbreak that slipped past the industry's strongest safeguards, the Cold War trade law repurposed to ground a chatbot, the Pentagon's move to blacklist the one AI company that drew red lines, and the foreign rival running twenty-five thousand fake accounts to copy the model's mind. Beneath the drama runs a single question, sharpening with every chapter: we have built machines that can work on their own, at scale - and we are still improvising the rules for what happens when we let them run.
Reported like a thriller and attributed like an investigation, this is the definitive account of the moment AI capability outran the institutions built to govern it.