What the Emails Say
On July 2, the US District Court for the Northern District of California unsealed a batch of correspondence, first reported by The Wall Street Journal, with Gizmodo subsequently publishing the full text. The two parties in the emails are Pentagon undersecretary for research and engineering Emil Michael (a former Uber executive) and Anthropic CEO Dario Amodei. The disagreement is clear: the Defense Department demanded Claude be usable for "all lawful purposes"; Amodei held firm on two points—fully autonomous weapons and domestic mass surveillance. In January this year, Michael offered a near-ultimatum in an email: Anthropic had "one last chance to align on core principles."
Blacklisted the Day After "Very Close"
The emails show Michael once called the two sides "very close" on contract terms in one email, but the very next day the Pentagon's supply-chain-risk determination on Anthropic took effect—at the time Anthropic had not yet been informed. Separately, per The Lever and ProPublica citing financial disclosures, Michael holds $2 million to $10 million in Perplexity shares and previously served on that company's board, and the doubts raised by this conflict of interest have received no official response so far, with the related allegations unverified.
The Red-Line Fight Goes Public
Amodei had said the same thing before in a CBS interview: AI systems are "far from reliable enough" to be entrusted with autonomous combat. This batch of emails turns the debate over the military boundaries of frontier models from statements of principle into a verifiable negotiation record—and so far, the only company to publicly say "no" happens to be the one placed on the risk list.