OpenAI's Pentagon Deal Exposes the Gap Between Its Stated Principles and Commercial Reality

A detailed analysis published Friday by The Register laid out what it described as a fundamental contradiction in OpenAI's handling of the Pentagon deal controversy. According to the piece, CEO Sam Altman had circulated an internal memo on the morning of February 27 articulating the same red lines Anthropic had drawn — no mass surveillance, no autonomous lethal weapons, humans remaining in the loop for high-stakes decisions.

Hours later that evening, OpenAI signed a deal with the Department of Defense that, in the words of a source cited by The Verge, boils down to: "If it's technically legal, then the US military can use OpenAI's technology to carry it out."

The Register noted that the deal allowed bulk data collection on Americans and placed no specific AI-based safeguards on military applications. Altman subsequently told employees that OpenAI had "no call on what the Department of War does with its AI." The analysis referenced OpenAI's mounting financial pressure — a burn rate exceeding $9 billion in 2025 — as context for why the company may have found the $200 million contract difficult to refuse.

The piece also pointed to market data showing OpenAI's enterprise share had declined from roughly 50% to 27% in 2025, while Anthropic had grown to 40% of enterprise large-language-model spend, according to Menlo Ventures. The Register argued the deal handed Anthropic a durable reputational advantage — particularly in European markets and among civil liberties-conscious enterprise buyers — that would be difficult to claw back.

The controversy also reinforced a structural question the episode raised about AI governance: at what point do vendor-imposed restrictions on how governments can use AI constitute legitimate ethical boundaries, versus interference with a client's sovereign operational authority?

Read the original reporting at The Register.