OpenAI Inks Classified DOD Deal

The deal, valued at up to $200 million, is managed by the Pentagon's Chief Digital and Artificial Intelligence Office (CDAO). Of the total amount, nearly $2 million in research, development, test, and evaluation funds were obligated at the time of the award for the initial phase. The project is expected to be completed by July 2026 and will be primarily carried out in the Washington, D.C. area. This agreement marks a significant shift in OpenAI's policy, which until January 2024 explicitly banned "military and warfare" applications from its usage terms. The company has since moved to a contract-based safety approach, allowing for defense work within specific constraints, a change that has been described as a "historic pivot." This new direction is part of a larger "OpenAI for Government" initiative, aimed at providing AI tools to various U.S. government agencies. The deal was finalized shortly after the Department of Defense's negotiations with rival AI company Anthropic broke down. Anthropic had insisted on contractual prohibitions against using its AI for mass surveillance and autonomous weapons, which the Pentagon would not agree to, leading to the government designating Anthropic a "supply-chain risk." OpenAI's CEO, Sam Altman, stated the company was able to reach an agreement because it was comfortable relying on existing laws and policies as safeguards. In response to public backlash and criticism that the initial agreement was "opportunistic and sloppy," OpenAI amended the contract. The new clauses explicitly ban the use of its AI for the intentional domestic surveillance of U.S. citizens, including through commercially acquired data, and prohibit intelligence agencies like the NSA from using its services without a separate agreement. Despite the inclusion of "red lines"—no mass domestic surveillance, no directing autonomous weapons, and no high-stakes automated decisions—critics point to the "all lawful purposes" language in the contract. They argue this phrasing creates a loophole, as the prohibitions are tied to current laws and policies which could potentially be changed by the government. The collaboration is not limited to software. Cleared OpenAI personnel will be directly involved to oversee the use of the models, which will be deployed via the cloud, allowing OpenAI to retain control over its safety systems. The project's scope focuses on transforming administrative operations, such as healthcare for service members, streamlining data analysis, and supporting proactive cyber defense. Looking ahead, the DOD plans to convene a working group that will include leaders from OpenAI and other AI labs and cloud providers. This group will serve as a forum for ongoing dialogue about emerging AI capabilities and national security challenges. This contract is seen as a test case for future federal AI procurement and governance frameworks.