The Pentagon is formalizing a new model for how frontier artificial intelligence systems enter classified military networks.
Under a newly reached agreement, OpenAI will deploy advanced AI models into classified environments while retaining technical and contractual safeguards.
The systems will be deployed exclusively via cloud-based infrastructure, allowing OpenAI to retain control over its safety stack, monitoring mechanisms, and enforcement controls.
OpenAI has said it will not provide edge-deployment versions of its models outside its operational oversight.
Instead, the models will run within a controlled framework that includes contractual safeguards, continuous monitoring, and the involvement of cleared company personnel supporting classified operations.
The agreement is governed by three stated limits: prohibiting mass domestic surveillance, preventing independent direction of autonomous weapons, and restricting high-stakes operational decisions without human control.
While specific missions were not disclosed, the structure aligns with the Pentagon’s broader push to integrate advanced AI into intelligence analysis, operational planning, and decision-support functions across classified networks.
The deal comes amid growing scrutiny of commercial AI use in defense systems, including the Trump administration’s directive ordering federal agencies to phase out Anthropic’s Claude model.
