TL;DR
Anthropic CEO Dario Amodei is making a final effort to salvage a $200 million Pentagon deal that collapsed over disagreements about autonomous weapons and mass domestic surveillance. He is now in direct talks with under-secretary of defence Emil Michael, after previously accusing the Pentagon and OpenAI of spreading falsehoods about the breakdown.
What Happened
Anthropic originally reached a $200 million agreement with the US Department of Defense in July 2025 to supply AI capabilities. That deal fell apart when both sides failed to agree on contract language concerning mass domestic surveillance and the use of AI in autonomous weapons systems.
The breakdown was anything but quiet. In a memo that surfaced publicly, Amodei accused the Pentagon and OpenAI of “lies” about the reasons for the collapsed deal. He also claimed Anthropic had been frozen out of defence contracting because the company refused to give “dictator-style praise to Trump.”
Defence Secretary Pete Hegseth escalated the standoff further by threatening to designate Anthropic as a supply chain risk — a move that would effectively bar the company from future government contracts and send a strong signal to other AI firms about the consequences of resisting Pentagon demands.
The New Talks
Amodei is now attempting to break the deadlock through direct discussions with Emil Michael, who serves as under-secretary of defence. The talks represent the most senior-level engagement between Anthropic and the Pentagon since negotiations broke down.
The stakes extend well beyond a single contract. For Anthropic, being shut out of US defence work while rival OpenAI moves in would represent a significant competitive setback. For the Pentagon, working with only one major AI supplier raises questions about vendor lock-in and the diversity of its technology base.
The situation also raises broader questions about the terms under which commercial AI companies engage with military and intelligence agencies. Anthropic’s original objections centred on specific use cases — not on working with defence altogether — suggesting there may be room for compromise if both sides can agree on acceptable guardrails.
Looking Forward
The outcome of these talks will shape the relationship between the US government and the AI industry for years to come. If Amodei can negotiate terms that preserve Anthropic’s safety commitments while satisfying Pentagon requirements, it could establish a template for responsible military AI procurement. If talks fail again, expect the divide between AI companies willing to work with defence on any terms and those insisting on restrictions to widen further — with direct implications for UK defence procurement decisions and NATO interoperability.