Google Signs Classified Pentagon AI Deal for 'Any Lawful Government Purpose' With No Veto Clause
Google has signed a classified agreement with the US Department of Defense allowing its AI models to be used for “any lawful government purpose,” including classified intelligence and military applications. The deal, first reported by The Information on April 28, contains no clause giving Google the right to veto or restrict specific government uses of its models.
The absence of a veto provision is the structurally significant detail. Google has previously pushed for use-case restrictions in enterprise AI agreements, and its 2018 decision to exit Project Maven was driven in part by a position that the company should retain oversight over military AI applications. That position does not appear in this contract.
The Terms
The classified nature of the agreement limits what can be confirmed publicly. What has been reported: the DoD can use Google’s AI models for “any lawful government purpose,” the scope covers classified environments, and Google does not retain a review or approval right over specific deployments.
The deal follows a pattern across the industry. Microsoft signed an agreement with the US military for Azure Government for classified workloads years ago. AWS GovCloud has hosted classified DoD AI work for years. OpenAI’s GPT-5.4-Cyber launched in April 2026 specifically targeting defense and intelligence applications. Google was, until now, the largest lab without a comparable classified-environment AI agreement.
Employee Backlash
Internal opposition to Google’s defense AI work has been a recurring issue since Project Maven in 2018. Employees circulated petitions opposing military AI contracts, and dozens resigned before Google did not renew the Maven contract.
The new DoD deal was finalised despite reported internal resistance, according to sources cited by The Information and The Verge. The current agreement appears broader in scope than Maven: Maven was a computer vision contract for drone surveillance; this deal covers AI model access across whatever lawful purposes the DoD elects to apply it to.
Strategic Context
Google DeepMind’s AI capabilities, particularly in reasoning and code, have become relevant to national security applications. Gemini 3.1 Pro Preview ties GPT-5.4 on the Artificial Analysis Intelligence Index at score 57. The US government has been explicitly trying to ensure American labs rather than Chinese competitors supply AI to defense and intelligence agencies.
The timing follows the White House’s release of a National AI Legislative Framework, the NSA’s deployment of Anthropic Mythos for cyber work, and OpenAI’s launch of GPT-5.4-Cyber. Defense is now a formal distribution channel for frontier AI, and Google is joining the others in serving it.
Whether the lack of a veto clause represents a policy shift or a negotiating outcome will likely remain unclear given the classified nature of the agreement.