The phrase that ended the Pentagon and Anthropic relationship was not in the message headers or the signatures. It sat in plain text inside an email from a Defense Department official: "all lawful uses." When Anthropic CEO Dario Amodei read the phrase, he recognized that current U.S. law already permits domestic surveillance, so a "lawful uses" guarantee could not function as a guardrail against the use cases his lab had spent years refusing to serve. The 346-page court filing now public captures how that single clause finished the conversation.
The unsealed documents, preserved at the Internet Archive, show a January 2026 email exchange between Amodei and Emil Michael, the Defense Department's Undersecretary of Defense for Research and Engineering and a former Uber executive, summarized in full this week by Gizmodo. After several weeks of radio silence, Michael wrote that he was "hoping that we are closer to engaging with your revised POV." He then laid out the Pentagon's "all lawful uses" standard as the floor for any deployment. Amodei's two redlines were already fixed in writing: Anthropic models would not be put inside fully autonomous weapons, and they would not power domestic surveillance tools.
Michael told Amodei that Anthropic's conditions were "just not workable" and offered "one more chance to align on core principles that would lead to legal language." He went further in his own words: "there is no distinction in our world between weapons that are defensive or offensive." Amodei answered in writing that the Pentagon's proposed language appeared to "completely remove our redlines," pointing out that "lawful" cannot exclude domestic surveillance in a country where the law itself permits it. Both sides used the same word, "redlines," but each meant the positions the other refused to accept.
The day after the failed exchange, Defense Secretary Pete Hegseth announced that the Department would designate Anthropic a supply-chain risk, a punitive procurement category that effectively blacklists a company from defense work. Subsequent reporting in Politico characterized Hegseth's ultimatum to Anthropic as "incoherent" to AI policymakers. Anthropic published its own official account through a statement attributed to Amodei, framing the dispute as a refusal to soften the lab's published use policy under procurement pressure.
The mechanism matters because the clause in question is not unique to Anthropic. "All lawful uses" is the Defense Department's standing procurement language, and every frontier AI lab that publishes ethical redlines while pursuing federal contracts will face the same collision. The Department of Defense's "no distinction" line on defensive versus offensive weapons, paired with U.S. law's permissive posture on domestic surveillance, leaves vendors with two options: sign the language and lose the policy they have spent years publishing, or refuse and absorb the procurement consequence. Reporting by The Guardian in April 2026 added a further dimension, linking Michael to millions of dollars in xAI stock sales after the Pentagon entered a separate agreement with xAI, raising a conflict-of-interest question that the Anthropic email exchange does not address.
The question now is whether other frontier labs treat the unsealed record as a usable case study or as a warning. Anthropic's earlier response after the supply-chain risk designation signaled the company would not quietly absorb the result. The Department of Defense has not, in the released filings, offered alternative language to close the gap. Watch whether the next AI vendor contract adds the missing distinction between defensive and offensive weapons, whether "all lawful uses" continues to travel as written, and how closely the email lines making the news actually track the original 346-page court filing, since the quoted exchanges have so far traveled through summary reporting rather than directly from the archive copy.