Andy Jassy didn't call a lawyer. He called the Treasury Secretary.
When Amazon researchers found a jailbreak in two of Anthropic's most powerful AI models — a vulnerability that could, in the wrong hands, strip away the guardrails designed to prevent misuse — the Amazon CEO picked up the phone and alerted Scott Bessent directly. That call, reported by the Wall Street Journal, set in motion the export controls issued on Friday that effectively shut down the models for international access. Jassy didn't leak it. He didn't sit on it. He went straight to the top of the US government.
The move was, by any reasonable measure, extraordinary. Amazon has committed billions to Anthropic. In return, Anthropic has pledged $100 billion in cloud spending back to AWS. The two companies are bound by one of the most significant commercial arrangements in the history of artificial intelligence. And Jassy just told the government about a flaw in the thing he helped fund.
That decision — and the architecture of relationships that made it so complicated — is what this story is really about.
