Anthropic statement on discussions with the Department of War
Key Points
- Two explicit exclusions: mass domestic surveillance and fully autonomous weapons
- Anthropic refuses to accept an "any lawful use" clause and won’t remove safeguards
- Will assist smooth transition if offboarded and remains open to R&D collaboration
Summary
Dario Amodei explains Anthropic's long-standing deployment of Claude across the Department of War and other national security agencies, describes two contractual safeguards the company will not remove, and outlines the operational and contractual implications if the Department insists on an "any lawful use" clause. The statement clarifies Anthropic's position on allowed and disallowed uses, past defensive actions against adversary misuse, and its readiness to support a transition if offboarded.
Key Points
- Anthropic has deployed Claude in classified networks, National Labs, and across multiple national security missions (intelligence analysis, modeling/simulation, operational planning, cyber operations).
- Company actions cited: blocked access for firms linked to the Chinese Communist Party, shut down CCP-sponsored abuse attempts, and supported export controls on chips—choices that cost near-term revenue.
- Two categorical exclusions from contracts: mass domestic surveillance and provision of systems to enable fully autonomous weapons (no humans in the loop).
- Rationale: mass domestic surveillance undermines democratic values and current law lags behind AI capabilities; frontier models are not sufficiently reliable today to safely power fully autonomous weapons.
- Anthropic offered to collaborate on R&D to improve reliability for future weapon use cases; the Department has not accepted that offer.
- The Department has sought contractors that accept "any lawful use" and has threatened to remove Anthropic from systems, designate the company a "supply chain risk," or invoke the Defense Production Act—actions Anthropic calls contradictory.
- Anthropic will not remove the two safeguards; it prefers to continue supporting the Department with those safeguards in place and will assist a smooth transition if the Department offboards Claude to avoid mission disruption.
- Models will remain available under the terms Anthropic has proposed for as long as required and the company remains ready to support U.S. national security.
Implications for engineers and program teams
- Contracts: expect negotiation friction if procurement requires an "any lawful use" commitment—clarify explicit allowed/disallowed use cases and verification mechanisms.
- Integration: plan for continuity and handoff processes if a provider change is required; expect potential administrative actions (designation, DPA invocation) that can affect timelines.
- Risk management: treat mass domestic-surveillance and fully autonomous-weapons use cases as disallowed by Anthropic today; design system requirements and architectures accordingly.
- R&D opportunities: Anthropic is willing to collaborate on improving reliability for high-assurance autonomy—teams should consider defining technical requirements, testing protocols, and guardrails for future work.