The US Department of Defense's relationships with the leading AI companies are far more intimate than either side has publicly acknowledged, according to more than 400 pages of contract documents obtained by The Intercept through a Freedom of Information Act lawsuit conducted with the group Legal Advocates for Safe Science and Technology. Reporter Sam Biddle published the documents on September 8, revealing details of the deals the Pentagon signed in July 2025 with Anthropic, Google, OpenAI and xAI — agreements, many of them heavily redacted, along with later amendments to the original terms. Together they provide the most concrete picture yet of how the Pentagon's flagship frontier AI procurement actually works, and of how little of it was visible in the companies' public announcements at the time.
What the contracts cover For more context on this story, see our ongoing AI trends.
Each of the four agreements is worth up to 200 million dollars — as much as 800 million dollars in combined potential value across the four companies — and commits the contractors to building prototypes of AI tools intended, in the contracts' language, to "improve military advantage, military utility, or enhance military decision making" across the armed forces. The applications span "warfighting," automated decision-making, logistics and military intelligence. According to The Intercept, these prototypes went on to form the basis of AI tools now deployed across the Department of Defense's classified networks — and the document set includes one contract reflecting OpenAI's classified network deployment, with equivalent documents for the other contractors still awaited.
A two-way exchange, not a software sale
The contracts describe arrangements that go well beyond selling access to models. An "information sharing" provision present in all four agreements creates a two-way exchange of data and expertise. The companies agreed not only to provide the Pentagon access to their best large language models but also to "inform DOD and CDAO strategy on frontier AI," as the Google agreement phrases it — with CDAO referring to the Defense Department's Chief Digital and Artificial Intelligence Office. In return, the documents show the AI labs were approved to receive an array of sensitive defense information.
That structure — labs shaping strategy while absorbing internal defense data — is what distinguishes the arrangements from ordinary cloud or software procurement. The Intercept's reporting emphasizes that the AI firms themselves are expected to influence how their technology is used to augment US capacity to wage war, rather than delivering a finished product and walking away.
The Anthropic rupture, revisited
Much of the public attention to the military AI deals has centered on Anthropic, whose falling out with the Pentagon earlier this year played out in pointed public statements about the ethics of AI in warfare and corporate power. The Intercept's reporting argues the "acrimonious breakup" actually hinged on a government contract that had remained hidden, and that the newly released paperwork illuminates the deals that led to the crisis. In a companion story published the same day, The Intercept reports that the Pentagon asked OpenAI for artificial intelligence designed to rarely refuse requests — a headline that captures the direction of the procurement demands described across the document set.
Experts flag surveillance and targeting
Sophia Goodfriend, a University of Cambridge research fellow and non-resident fellow at the Harvard Kennedy School's Middle East Initiative who studies machine learning and military conflict, told The Intercept the documents appear to be "the first time the extent of collaboration between frontier AI labs and the Pentagon is spelled out in explicit terms." What is particularly notable, she said, are "the terms by which engineers are working in lockstep with the department of war to engineer AI systems for surveillance, targeting, and killing."
Official statements versus contract reality
When the deals were announced in July 2025, public statements were light on specifics. Pentagon spokesperson Sean Parnell said at the time, "We must equip our warfighters with 21st-century technology in order to defeat 21st-century threats." Google described "advanced AI solutions" that would let the Defense Department "scale the adoption of agentic AI across enterprise systems." The contract paperwork now shows how much those bland descriptions left out: engineers embedded in iterative prototyping for combat-relevant tasks, strategy-level influence over defense AI planning, and deployments that reached classified networks.
What to watch
The release raises immediate oversight questions. The documents obtained so far are heavily redacted, and The Intercept is still awaiting the government's production of equivalent contract documents for Anthropic, Google and xAI — meaning the fullest picture of the military's frontier AI partnerships has yet to emerge. What is already public gives lawmakers a concrete starting point: the actual contract terms, amendments and deliverables behind tools that are already operating on classified systems. How the four companies reconcile their public safety commitments with contract obligations to the Pentagon — and whether Congress demands the same disclosure for the remaining agreements — will shape the next phase of the military AI debate.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
