What happened: Anthropic CEO Dario Amodei published a full statement on the company's dispute with the Department of War, laying out in detail why Anthropic refuses to remove two specific safeguards from its contracts: a prohibition on use for mass domestic surveillance and a bar on powering fully autonomous weapons systems. He stated the company "cannot in good conscience" comply with the Pentagon's demands, but expressed a strong preference to keep serving the Department — with the safeguards intact.

Why it matters: The statement reveals the breadth of Anthropic's existing military deployment: Claude is already used across DoD classified networks, national laboratories, and national security agencies for intelligence analysis, operational planning, and cyber operations. Amodei argues neither of the contested safeguards has been a barrier to any of that existing work. The DoD's stated position — that it will only contract with companies agreeing to "any lawful use" and removing all safeguards — amounts to demanding companies abandon their own safety policies as a condition of doing government business.

Wider context: Amodei notes the DoD's twin threats are "inherently contradictory": designating Anthropic a supply chain risk — a label previously reserved for US adversaries, never before applied to an American company — while simultaneously invoking the Defense Production Act on the basis that Claude is essential to national security. He also disclosed that Anthropic forewent several hundred million dollars in revenue by cutting off CCP-linked firms, and was the first frontier AI company to deploy models on US classified networks.

Background: The two contested uses: first, AI-powered mass domestic surveillance — which Amodei argues is currently legal only because law hasn't caught up with AI's ability to assemble individually innocuous public data into comprehensive profiles at scale. Second, fully autonomous weapons — which he says today's AI is not reliable enough to operate safely. Anthropic offered to work with the DoD on R&D to improve that reliability. The Department has not accepted the offer.

Statement from Dario Amodei on our discussions with the Department of War — Anthropic

Singularity Soup Take: Anthropic's statement is a rare thing in corporate America — a company telling the government "no" in writing, on principle, with receipts — and whatever you think of the outcome, the clarity of the reasoning makes it worth reading in full.

Key Takeaways:

• Already Deep In: Claude is deployed across DoD classified networks, national laboratories, and national security agencies — Amodei says the two contested safeguards have not been a barrier to any existing military use to date.
• "Any Lawful Use" Demand: The DoD's stated contracting policy requires AI companies to agree to remove safeguards for any lawful purpose — a blanket condition Anthropic says it cannot accept without compromising its core principles.
• Contradictory Threats: The DoD simultaneously threatened to label Anthropic a "supply chain risk" (a designation previously reserved for US adversaries) and to invoke the Defense Production Act to compel compliance — two positions that directly contradict each other.
• Revenue Sacrifice: Anthropic cut off CCP-linked firms at a cost of several hundred million dollars in revenue — a decision Amodei cites as evidence of genuine commitment to US national security interests, not obstruction of them.
• R&D Offer Rejected: Anthropic offered to collaborate directly with the DoD on improving the reliability of autonomous weapons systems — an offer the Department has not accepted.

Related News

Anthropic CEO Refuses Pentagon Demand to Remove AI Safety Limits — Earlier coverage of the same dispute, based on reporting from the BBC.

Anthropic Scraps Its Core Safety Promise Amid Pentagon Pressure — Background on how DoD negotiations began reshaping Anthropic's public commitments.