U.S. Department of War

The Department of War integrated OpenAI's AI models into classified military networks under an agreement reached in February 2026, allowing the use of advanced AI for both warfighting and enterprise tasks in secure, cloud-based classified environments. The agreement includes contractual limits prohibiting use for mass domestic surveillance or directing autonomous weapons systems.

On February 27, 2026, OpenAI and the Department of War reached a formal agreement for deploying advanced AI systems in classified environments. OpenAI's models are deployed via cloud-only infrastructure using OpenAI's safety stack; the company is not providing 'guardrails off' or non-safety-trained models, and edge deployment is excluded. Cleared, forward-deployed OpenAI engineers and safety researchers remain involved in oversight. Permitted uses, as confirmed in a separate OpenAI announcement about the GenAI.mil platform, include summarizing and analyzing policy documents, drafting procurement materials, generating internal reports, and supporting research, mission planning, and administrative workflows. The agreement explicitly bars use of the AI for domestic surveillance of U.S. persons (including through commercially acquired data), directing autonomous weapons, or high-stakes automated decisions. The deal followed the collapse of a similar agreement with Anthropic; on February 27, 2026, the Department designated Anthropic a supply-chain risk to national security.

The Department of War announced an Artificial Intelligence Acceleration Strategy on January 9, 2026, directing the department to develop AI agents for battle management and decision support, mandate federated data access for AI systems, and require that the latest commercial AI models be made available within 30 days of public release. By April 2026, DoW personnel had used GenAI.mil's Agent Designer tool to build over 100,000 agents for unclassified administrative tasks, while the strategy's Agent Network initiative continues to direct AI agent development for battle management and decision support.

The January 9, 2026 strategy memo directs the CDAO to oversee seven Pace-Setting Projects (PSPs) spanning warfighting, intelligence, and enterprise mission areas. One PSP, called Agent Network, directs AI agent development and experimentation for battle management and decision support, from campaign planning through targeting. The memo requires all military departments to deliver federated data catalogs to the CDAO within 30 days and mandates that AI model contracts include the latest model versions within 30 days of public release. The AI Rapid Capabilities Cell (AI RCC), formed in December 2024 with $100 million in FY2024-2025 funding and run jointly by the CDAO and the Defense Innovation Unit, is executing frontier AI pilots across warfighting and enterprise use cases. Separately, by April 2026 DoW personnel had built over 100,000 agents on GenAI.mil using a no-/low-code Agent Designer tool for unclassified administrative tasks such as drafting documents and summarizing meetings; the platform had more than 1.2 million discrete users as of that date.

The Department of War deployed the Maven Smart System (MSS), an AI platform built by Palantir, to analyze data from drones, satellites, radar, and other sensors to identify objects and potential targets and support battlefield decision-making. The program traces its roots to Project Maven, launched in 2017; Palantir became the primary contractor under a May 2024 contract. As of 2026, the Pentagon is expanding MSS department-wide as a formal program of record.

Maven Smart System grew out of Project Maven, launched in 2017, which initially applied computer vision to analyze drone footage. The system processes inputs including satellite imagery, full-motion video, geolocation data, and radar to detect, classify, and track objects and personnel of interest, and surfaces potential targets to human analysts for review. It also includes an AI Asset Tasking Recommender that can suggest which weapons should be assigned to which targets, with human officers retaining final authority. In April 2026, the Pentagon requested $2.3 billion over five years to expand Maven, and the system has been described as the Department's primary battlefield command-and-control, targeting, and situational awareness platform.