Alphabet’s Google has entered into an agreement with the U.S. Department of Defense to permit use of the company’s artificial intelligence models in classified work, a person familiar with the situation said on Tuesday.
Under the terms described by that person, the Pentagon will be authorized to use Google’s AI models for "any lawful government purpose." The description of permitted uses is explicit in scope, according to the person briefed on the arrangement.
The Defense Department has previously executed comparable arrangements with other prominent AI developers, including OpenAI and xAI, enabling those firms' models to be utilized in classified settings.
The deal with Google arrives after a notable disagreement between the Defense Department and Anthropic, the developer of the Claude model, over the permissible military and surveillance applications of that firm’s AI.
Separately, a media report published on Monday indicated that hundreds of Google employees have petitioned the company’s chief executive, Sundar Pichai, urging him to decline classified AI assignments with the Pentagon. The petition, according to that report, represents internal concern among staff over the company’s participation in classified government work.
Taken together, the disclosed deal and the surrounding developments highlight two concurrent dynamics: the Defense Department’s ongoing efforts to secure access to advanced AI capabilities from major technology firms, and the internal and intercompany frictions that can arise when AI technology is applied to military or surveillance tasks.
Details available in the public reporting do not enumerate specific operational projects, contract values, or timelines tied to Google’s arrangement with the Pentagon. The person familiar with the matter provided the characterization of permitted uses but did not disclose additional contractual terms.
Observers and market participants will likely monitor how this agreement, and related industry disputes and employee actions, evolve. For now, the disclosed facts are limited to the existence of the agreement, the stated permissive language regarding lawful government purposes, prior comparable deals with other AI firms, the Anthropic-DoD disagreement, and the reported internal petition at Google.