A leaked internal memo from Anthropic CEO Dario Amodei has pushed a private fight over military AI into public view. In the note, Amodei called statements from OpenAI CEO Sam Altman “a complete lie�� and described OpenAI’s latest Pentagon arrangement as “security theater.” The dispute is no longer about messaging alone. It is now a direct clash over how far AI companies should go in defense work and what limits must stay in place.
A $200 million defense contract set off the break
The immediate trigger was Anthropic’s existing $200 million defense contract. Through a partnership with Palantir, Claude had already been deployed on classified military networks. By late February, the Pentagon gave Anthropic an ultimatum: remove all AI usage restrictions and allow unrestricted access for “any lawful purpose,” or lose the contract by February 27 and face blacklisting.
Amodei rejected the demand in public. He said he could not accept those terms in good conscience and held to two red lines: no autonomous weapons systems making final battlefield targeting decisions, and no tools for mass domestic surveillance of U.S. citizens. The response came quickly. According to the source material, within hours of Anthropic’s refusal, the Trump administration labeled the company a “supply chain risk,” a move that effectively shut it out of federal contracts.
OpenAI accepted a broader legal-use standard
On February 28, only hours after Anthropic was cut off, Altman announced that OpenAI had reached an agreement with the Department of War. OpenAI said in a blog post that the contract carried protections similar to Anthropic’s stated red lines, including limits on autonomous weapons, mass domestic surveillance, and critical automated decision-making.
The gap sits in the wording. Anthropic wanted explicit bans written into the deal. OpenAI accepted language allowing “all lawful uses.” OpenAI said the Department of War had made clear during discussions that mass domestic surveillance was illegal and not part of its plans. Critics focused on a simple problem: laws can change. Conduct treated as illegal now may be allowed later, which leaves a contract built around “lawful use” on shakier ground than one with direct prohibitions.
The leaked memo exposed the real argument
Amodei’s memo was blunt about the public fallout. He wrote that most people and much of the media saw OpenAI’s deal with the Department of War as suspicious, while viewing Anthropic as the heroic side in the dispute. It was a sharp assessment. It was also a very public one once the memo leaked.
TechCrunch reported that Amodei also accused Altman of presenting himself as a peacemaker and master dealmaker. He argued that the main difference between the two companies was motive: OpenAI cared about calming employees, while Anthropic cared about preventing misuse. Altman, for his part, acknowledged in an OpenAI all-hands meeting that the decision had created severe brand consequences, but he still defended it as a difficult yet correct choice for national security.
User behavior is already shifting
The backlash is not limited to internal debate or press coverage. The source says ChatGPT uninstall numbers have surged, while downloads of Anthropic’s Claude app have risen sharply. No exact figures were provided, but the direction of user response is clear enough.
Each company is carrying a different cost. Anthropic refused the Pentagon’s terms and lost federal business along with government ties. OpenAI chose to work within the contract while keeping stated limits, and the price has fallen on brand trust and user confidence. The split has also exposed a larger fault line: as AI moves deeper into military systems, the distance between what is legal and what is right is becoming harder to ignore.