Unsealed Emails Reveal Pentagon Pressed Anthropic to Drop Autonomous-Weapons Red Lines

Court documents unsealed on July 2 in the US District Court for the Northern District of California — first reported by the Wall Street Journal — reveal the private negotiation that preceded the Pentagon's decision to designate Anthropic a "supply-chain risk," and the two red lines the company refused to cross.

The emails, exchanged over several months between Anthropic CEO Dario Amodei and Emil Michael, the Pentagon's Under Secretary of Defense for Research and Engineering, show the dispute was never about whether the military could use Claude. It was about control: whether the Department of Defense could deploy the model for fully autonomous weapons and domestic surveillance without restriction.

The Two Red Lines

Amodei held to two conditions throughout the talks: Claude could not be used in fully autonomous weapons systems, and it could not be used for domestic surveillance of Americans. When the Pentagon's proposed contract language appeared to "completely remove our redlines," as Amodei wrote, Michael responded that Anthropic's guardrails were "just not workable."

The talks began to sour in January, the documents show, when Michael emailed Amodei after weeks of silence saying he was "hoping that we are closer to engaging with your revised POV." Subsequent exchanges grew increasingly terse as contract drafts went back and forth.

A Striking Timeline

Perhaps the most consequential revelation is chronological. Michael emailed Amodei declaring the two sides "very close" on contract terms — one day after the supply-chain risk designation had already been finalized internally, and before Anthropic had been informed. Defense Secretary Pete Hegseth announced the designation publicly the following day.

Anthropic's lawsuit argues the designation was retaliation for refusing the Pentagon's terms rather than a genuine security assessment. Amodei has said publicly that AI systems are "nowhere near reliable" enough to make lethal decisions autonomously, and that the company will accept lost revenue rather than concede the point.

Industry Fault Line

The unsealed correspondence lands amid an unprecedented show of cross-company solidarity: employees at OpenAI and Google have signed open letters backing Anthropic's position in the litigation. With Google's Gemini expanding inside the Pentagon's GenAI.mil program, the documents sharpen the industry's central ethical divide — between labs that treat military guardrails as negotiable contract terms and one that has now gone to federal court to defend them.