Anthropic CEO Calls OpenAI’s Military Messaging ‘Straight Up Lies’

Anthropic CEO Dario Amodei has launched an unusually direct public attack on rival OpenAI, accusing the company of misleading the public about its defense work through what he called 'straight up lies' and 'safety theater.' The dispute, reported by TechCrunch via The Information, stems from a leaked internal memo where Amodei contrasted Anthropic's approach with OpenAI's. While Anthropic's talks with the Department of Defense for a potential $200 million contract reportedly collapsed after the Pentagon sought 'unrestricted access' without guarantees against mass domestic surveillance or autonomous weaponry, OpenAI proceeded to reach an agreement. Amodei framed the conflict as a fundamental question of where AI companies draw ethical lines and how honestly they describe those boundaries when dealing with the U.S. military.

The core technical and contractual flashpoint is OpenAI's use of the phrase 'all lawful purposes' in its public description of the defense deal, alongside stated red lines prohibiting mass surveillance, autonomous weapons, and high-stakes automated decisions. OpenAI argues these restrictions are explicitly written into the contract itself and paired with the 'lawful purposes' clause, positioning them as legally binding commitments rather than just blog post promises. The argument highlights a critical market question: when an AI vendor describes use restrictions, are those limits enforced through ironclad contract terms, technical controls, or merely public relations? As defense and enterprise customers demand more clarity, companies will face pressure to be more precise in defining what their models can and cannot do, with this public clash between two AI giants setting the stage for increased scrutiny over AI-military partnerships.