What are the terms of OpenAI’s deal with the Department of Defense?

Core points of the agreement

OpenAI reached an arrangement with the Department of Defense to let its models operate inside a classified DOD network, and company leaders say the contract preserves several safety commitments. The agreement was announced publicly by OpenAI’s CEO and framed as an effort to let the military use advanced models while keeping limits on certain applications.

OpenAI and the Pentagon described the deal as including technical safeguards and operational controls intended to prevent uses the company has publicly ruled out. Company statements stress that the models will not be deployed for fully autonomous lethal systems or for mass domestic surveillance. OpenAI characterized the DOD’s final terms as respecting those red lines and said the arrangement contains more guardrails than prior classified deployments by other providers.

How the pact came together and its broader context

• The deal followed a public fallout between the Pentagon and another AI company, raising pressure to find an alternative supplier.
• OpenAI has said the DOD will allow it to build a safety or “responsibility” stack for classified deployments, and the company asked the Defense Department to extend the same protections to other trusted AI firms.
• Details about auditing, data handling, or specific technical controls were not disclosed in full; the company described the protections at a high level.

Why this matters

Putting a mainstream model inside a classified environment is a precedent for integrating large language models into sensitive government workflows. If the safeguards prove robust, the arrangement could become a template for future classified AI use; if problems emerge, it will intensify scrutiny of how commercial AI and national security intersect.