What deal did OpenAI make with the Defense Department?

OpenAI cleared to run models inside a classified DOD environment

OpenAI announced an agreement with the Defense Department that allows its AI models to be deployed inside a classified DOD network. Company leadership said the arrangement includes safety and operational terms intended to let the agency use OpenAI’s systems in sensitive environments while imposing limits on misuse.

The timing was notable: the deal was announced amid a broader government rift with Anthropic, which refused to remove guardrails the Pentagon wanted. Reports indicate the DOD accepted safety proposals from OpenAI that resemble the red lines Anthropic had proposed; OpenAI also asked the department to apply the same terms to other AI vendors. Sam Altman publicly framed the agreement as a model the Pentagon could extend across industry.

Key elements and implications

• Deployments: OpenAI will run models in a classified environment separate from public cloud access, intended to meet the DOD’s handling rules for sensitive data.
• Safety architecture: The department reportedly permitted OpenAI to implement a tailored “safety stack,” a set of technical, policy, and operational controls the company will operate.
• Industry precedent: If the DOD treats OpenAI’s red lines as a template for other vendors, it could standardize how military customers contract with AI providers.

What remains unclear

• Technical details about the safety stack and specific prohibitions were not publicly disclosed.
• Whether the agreement will prevent future disputes over particular use cases—such as autonomous weapons or domestic surveillance—depends on how the DOD, Congress, and contractors define lawful uses going forward.

The deal marks a turning point in how national security customers and AI companies negotiate control over advanced models in classified settings.