OpenAI Agreement with the Department of War

OpenAI has reached an agreement with the Department of War (DoW) to deploy advanced AI systems within classified environments. This partnership is designed to provide the U.S. military with high-performance AI tools while maintaining strict technical and contractual guardrails to prevent misuse in domestic surveillance and autonomous weaponry.

Core Safety Red Lines

OpenAI has established three primary "red lines" that the Department of War is prohibited from crossing under the terms of the agreement:

  • Mass Domestic Surveillance: OpenAI technology cannot be used for mass domestic surveillance of U.S. persons.
  • Autonomous Weapons Systems: OpenAI technology cannot be used to direct autonomous weapons systems.
  • High-Stakes Automated Decisions: OpenAI technology cannot be used for high-stakes automated decisions, such as "social credit" systems.

Multi-Layered Enforcement Mechanism

To ensure these red lines are enforceable, OpenAI employs a layered security and deployment strategy rather than relying solely on usage policies.

Deployment Architecture

The system is a cloud-only deployment. OpenAI explicitly refuses to provide "guardrails off" or non-safety trained models and will not deploy models on edge devices. This architecture ensures that OpenAI can independently verify usage through the operation of classifiers and prevents the technology from being integrated into autonomous lethal weapons, which would require edge deployment.

Contractual Protections

The agreement includes specific legal language to restrict usage:

  • Autonomous Weapons: The AI system will not independently direct autonomous weapons where law, regulation, or Department policy requires human control. All AI use in autonomous systems must follow DoD Directive 3000.09 (January 25, 2023), requiring rigorous verification and testing.
  • Intelligence and Privacy: Handling of private information must comply with the Fourth Amendment, the National Security Act of 1947, the Foreign Intelligence Surveillance Act (FISA) of 1978, and Executive Order 12333. The system cannot be used for unconstrained monitoring of U.S. persons' private information or domestic law-enforcement activities, except as permitted by the Posse Comitatus Act.
  • Future-Proofing: The contract references laws and policies as they exist at the time of the agreement. This ensures that if laws or policies change in the future, the use of the systems must still align with the standards established in the current agreement.

Human Oversight

OpenAI will embed cleared forward-deployed engineers and safety and alignment researchers within the government's operations to provide additional assurance and oversight.

Clarifications on Domestic Surveillance

An update published on March 2, 2026, further clarified the agreement's stance on domestic surveillance. The updated language explicitly prohibits the use of OpenAI tools for domestic surveillance of U.S. persons and nationals, including the procurement or use of commercially acquired personal or identifiable information.

Additionally, the Department of War affirmed that OpenAI services will not be used by intelligence agencies such as the NSA; any such use would require a separate agreement.

Industry Context and Collaboration

OpenAI states that this agreement provides more responsible safeguards than previous agreements made by other labs, specifically citing Anthropic. OpenAI has requested that the Department of War make these same terms available to all AI companies to encourage a collaborative and stable relationship between the government and frontier AI labs.

As part of this initiative, the Department of War plans to convene a working group consisting of leaders from frontier AI labs, cloud providers, and the Department's policy and operational communities to discuss emerging AI capabilities, privacy, and national security challenges.

Sources