The U.S. Department of Defense has signed classified contracts with eight AI companies — excluding Anthropic — to supply AI systems for military use. Officials say the agreements formalize the Pentagon’s push to integrate advanced artificial intelligence into operational missions.
On May 1 (local time), the Pentagon announced it finalized deals with OpenAI, Google, NVIDIA, Microsoft, Amazon Web Services (AWS), Oracle, SpaceX and Reflection AI to permit their AI technologies to run on classified military networks.
The department said the contracts authorize use of AI models and infrastructure on classified and top‑secret networks under conditions of “lawful operational use.” That access is intended to expand AI support for intelligence analysis, mission planning, target detection and battlefield situational awareness.
Emil Michael, the Defense Department’s deputy for research and engineering, said the goal is to “equip warfighters with a range of AI tools to secure an overwhelming decision‑making advantage,” calling the move “a key step in transforming the U.S. military into an AI‑first fighting force.”
The agreements cover closed‑source large language models from OpenAI, Google and xAI, alongside open‑source offerings from NVIDIA and Reflection AI. Microsoft and AWS will supply AI infrastructure, while Oracle will provide cloud and AI platform services.
The deal with NVIDIA is notable for underscoring the Pentagon’s expanded open‑source AI strategy. NVIDIA will provide the open‑source model Nemotron, which the company says can support development of autonomous AI agents. CEO Jensen Huang has argued that open models — because their attributes are transparent and they can be tailored for specific missions — can enhance safety and security.
Anthropic’s exclusion drew particular attention. The company’s Claude model had been a major tool in classified environments through Palantir’s military AI platform, Maven, but the Pentagon designated Anthropic a supply‑chain risk after a dispute with the company.
Relations briefly appeared to ease when Anthropic visited the White House to discuss using its Misos model for cybersecurity capabilities. Still, Defense Secretary Pete Hegseth criticized Anthropic CEO Dario Amodei as an “ideological zealot” during a congressional hearing this week, signaling that the underlying dispute remains unresolved.
Pentagon officials emphasized they are avoiding dependence on any single company or AI model. “Diversity and flexibility in AI capabilities matter on the battlefield,” a Defense Department official said, adding that contracts with multiple firms create a portfolio approach that prevents technical lock‑in.
Concerns about expanding military use of AI persist. Civil society groups and some AI‑safety researchers warn that relying on AI for target selection or combat decisions could introduce errors and automation bias that distort human judgment.
Still, U.S. defense leaders view AI as a core competitive edge on future battlefields. Cameron Stanley, the Pentagon’s chief digital and AI officer, said, “AI‑enabled human‑machine teams will analyze vast amounts of data and grasp complex environments faster, helping commanders make better decisions.” He called the agreements a decisive turning point for securing the U.S. military’s next‑generation battlefield advantage.
Reporter Chan Park cpark@aitimes.com
