OpenAI has reached an agreement with the Department of War to deploy advanced AI systems in classified environments, establishing what the company claims are the strongest safety guardrails of any such contract. The deal includes three explicit redlines: no use for mass domestic surveillance, autonomous weapons, or high-stakes automated decisions like social credit scoring.

Unlike other AI labs that have reduced safety guardrails for national security work, OpenAI maintains full control over its safety stack and requires cleared OpenAI personnel to remain in the deployment loop. The agreement specifies cloud-only deployment architecture, preventing edge device usage that could enable autonomous lethal weapons. OpenAI has also requested that these terms be made available to all AI companies, including Anthropic, as part of broader collaboration efforts.

The contract explicitly references current U.S. laws and DoD policies, ensuring that even if regulations change in the future, AI systems must still operate within today's standards. OpenAI engineers will work directly with government personnel, with safety and alignment researchers maintaining oversight. The company frames this as essential for supporting U.S. military capabilities against growing threats from potential adversaries while maintaining democratic values and technological safeguards.