OpenAI, Google, Anthropic, and xAI to shape AI use in US Military #
Documents obtained by The Intercept through a Freedom of Information Act lawsuit reveal that OpenAI, Google, Anthropic and xAI each signed Pentagon contracts with ceilings of up to $200 million, agreeing to help shape how artificial intelligence is deployed in military decision-making, intelligence analysis and operational planning.
The contracts were awarded in July 2025, with subsequent documents covering amendments and agreements relating to deployments in classified military environments.
The more than 400 pages of documents, obtained with the help of Legal Advocates for Safe Science and Technology, cover the July 2025 deals and later contractual developments involving the companies and the Pentagon.
Companies Agreed To Help Shape Pentagon AI Strategy #
Under the agreements, the four firms committed to building prototype AI tools intended to 'improve military advantage, military utility, or enhance military decision making'.
The contracts contained an information-sharing clause requiring the companies to help 'inform DOD and CDAO strategy on frontier AI', indicating the deals went beyond a straightforward software sale.
The contracts also provided for two-way information sharing, with the companies receiving access to operational briefings, threat information and other Pentagon data.
Google was tasked with briefing the military on the 'AI tactics and techniques of adversaries' of the United States, according to the documents.
The firms also agreed to run 'tabletop exercises' and offer 'risk forecasting' on the dangers their own products might pose. Heidy Khlaaf, chief scientist at the AI Now Institute and a former OpenAI safety engineer, described the arrangement as troubling.
She argued that allowing companies to assess the risks of their own technology amounts to a 'subversion of democratic processes when AI labs are allowed to take over the arbitration of risk determinations with life-or-death consequences'.
Anthropic's Refusal Triggered a Pentagon Dispute #
The contracts also shed light on the collapse of Anthropic's relationship with the Department of Defense.
The company refused to agree to terms permitting unrestricted deployment on classified networks unless its restrictions on autonomous weapons and domestic surveillance were maintained, according to contractual amendments reviewed by The Intercept.
That dispute led Defense Secretary Pete Hegseth to designate Anthropic a 'supply chain risk' in late February, with the formal notification following in March.
A federal judge subsequently ruled that the Pentagon's designation of Anthropic as a supply-chain risk was unlawful. Despite the ruling, a senior Pentagon official said in September that the department continued to regard Anthropic as a supply-chain risk.
OpenAI, Google and xAI subsequently reached their own arrangements concerning classified-network deployments. OpenAI's agreement was announced on 28 February and provides for deployment of advanced AI systems in classified environments.
OpenAI said its agreement prohibits the use of its technology for mass domestic surveillance, directing autonomous weapons systems and certain high-stakes automated decisions.
Critics Warn of Conflict of Interest as AI Firms Assess Their Own Military Risks #
Sophia Goodfriend, a University of Cambridge research fellow, said the paperwork shows how closely 'engineers are working in lockstep with the department of war to engineer AI systems for surveillance, targeting, and killing'.
The implications extend beyond the four companies involved. Khlaaf said relying on AI companies to assess and report the risks of their own products creates a conflict of interest and argued that the public would be better served by independent assessments.
The Wall Street Journal has separately reported that Anthropic's Claude was used by US Central Command for intelligence assessments, target identification and battle simulations during US strikes on Iran.
Other reporting has also confirmed that Anthropic's AI services were used by the Pentagon during the operation.
The precise extent to which Claude directly influenced individual targeting decisions has not been publicly established.
OpenAI said its agreement bars use for mass domestic surveillance, autonomous weapons direction or high-stakes automated decisions. Anthropic, Google and xAI did not respond to requests for comment on their contracts.
OpenAI's published description of its agreement likewise says its technology cannot be used to direct autonomous weapons systems or for mass domestic surveillance.
These contracts provide some of the clearest public documentation yet of how deeply frontier AI companies are becoming involved in US military planning, despite the companies' stated restrictions on some military and surveillance uses of their technology.
As classified deployments expand, the documents are likely to fuel further scrutiny from lawmakers, researchers and rights groups over the relationship between companies' public safety commitments and their government contracts.
© Copyright IBTimes 2026. All rights reserved.