Leaked Emails Reveal Pentagon-Anthropic Dispute Over Autonomous Weapons

A federal court in California unsealed email correspondence between Anthropic CEO Dario Amodei and Emil Michael, the US Under Secretary of Defense for Research and Engineering. The emails show how a dispute over the limits of military use of Claude led to Anthropic being placed on a blacklist of suppliers deemed a risk to the supply chain.

The dispute has been running since January 2026, when Michael reached out to Amodei after weeks of silence, signaling a willingness to resume talks on the Department of Defense's continued use of Claude. The negotiations concerned the terms for deploying the model on GenAI.mil, the military's internal system for working with generative artificial intelligence.

At the Heart of the Dispute

From the start of talks, Amodei insisted on two restrictions: Claude could not be used to control weapons systems operating fully autonomously, without a human making the decision to use force in the moment, and it could not be used for mass surveillance of US citizens. The Pentagon wanted language that was much broader, covering "all lawful uses" - and since US law permits domestic surveillance under certain conditions, such wording would in practice have eliminated the restriction Anthropic cared about.

There is no distinction in our world between defensive and offensive weapons - Emil Michael, US Under Secretary of Defense

Responding to the Pentagon's proposal, Amodei wrote bluntly that the proposed wording "completely removes our red lines." Michael did not dispute that assessment, replying that the safeguards Anthropic was demanding were "simply unworkable."

That's simply unworkable - Emil Michael, US Under Secretary of Defense, on the safeguards demanded by Anthropic

Blacklist and Suspicious Timing

After the negotiations collapsed, Defense Secretary Pete Hegseth announced that Anthropic would be added to the supply-chain risk list, a classification normally reserved for companies linked to foreign adversaries, not an American software vendor. The decision was announced in February, and just one day after it was finalized, before Anthropic had even been formally notified, Michael emailed Amodei stating that the two sides were "very close" to an agreement on contract terms.

That discrepancy in timing is one of the key elements of the lawsuit filed by Anthropic. The company filed a civil suit in the District Court for the Northern District of California and an appeal in the Court of Appeals for the District of Columbia Circuit, challenging both the order barring federal agencies from using its products and the risk-list designation itself. In late March, a federal judge granted Anthropic a temporary restraining order, calling the administration's actions "a classic unlawful retaliation for exercising First Amendment rights." In April, however, the appeals court overturned that decision.

Anthropic's Position

Amodei has repeatedly stressed that he does not deny the future usefulness of autonomous combat systems for US defense, but considers the current reliability of models insufficient to hand them full control over the use of force. That distinction - between refusing to work with the military altogether and refusing to approve one specific, risky application - is at the center of the entire dispute, and sets Anthropic's position apart from the previously common assumption that the company avoids defense contracts as a matter of principle.

Industry Implications

The case shows that government contracts for AI models are no longer purely a matter of price and performance, but are becoming a battleground over who sets the limits on how the technology is used - the vendor or the customer. For software companies working with the defense sector in Poland and Europe, it's a signal that clauses on permissible AI model use in public contracts will increasingly be the subject of hard negotiation rather than a formality.

The US Congress has already announced it will take a closer look at the case, treating it as a precedent for future defense procurement of AI systems. The fate of Anthropic's Pentagon contract remains unresolved, and the legal dispute continues in parallel with attempts to resume business talks between the two sides.

Sources: The Next Web (thenextweb.com), Tech Times (techtimes.com), CNBC (cnbc.com)