Google Is Negotiating a Deal With the US DOD to Deploy Gemini AI in Classified Settings

Google Is Negotiating a Deal With the US DOD to Deploy Gemini AI in Classified Settings

Google is in negotiations with the US Department of Defense to allow the Pentagon to deploy Gemini AI models in classified settings, according to sources familiar with the discussions. The move would mark a significant reversal for Google, which famously walked away from Project Maven in 2018 following employee protests against military AI contracts. If completed, the deal would make Google one of the primary AI model providers for US defense agencies operating in sensitive, air-gapped environments.

What the Deal Would Cover

The negotiations center on allowing DOD agencies to run Gemini models on government-controlled infrastructure, including classified networks inaccessible to the public internet. This is a fundamentally different arrangement than cloud-based AI access: the Pentagon would be deploying Gemini within its own secure perimeter, giving it control over data sovereignty and preventing any information from flowing back to Google's commercial systems.

This model — sometimes called "government cloud" or classified deployment — is already used by Amazon Web Services and Microsoft Azure, both of which have established GovCloud environments serving intelligence and defense customers. Google Cloud has had a FedRAMP-authorized offering but has not previously extended AI models to classified tiers.

Reversing the Post-Maven Stance

Google's exit from Project Maven in 2018, driven by a petition signed by thousands of employees opposed to AI being used to improve drone targeting, shaped the company's posture on defense contracts for years. The company published AI Principles that included language about not deploying AI in weapons systems, and for several years maintained more cautious engagement with military customers than Microsoft or Amazon.

The shift under Sundar Pichai has been gradual. Google has renewed defense-adjacent contracts and expanded its FedRAMP offerings, but a direct DOD deal for classified AI deployment at this scale would be qualitatively different from previous steps. It would effectively put Google in the same category as Microsoft, which has deployed Azure OpenAI models for classified government use.

The Competitive Pressure

The decision is driven partly by competitive reality. Microsoft Azure Government and AWS GovCloud are deeply embedded in the defense and intelligence community. If Google stays out of classified AI deployments, it risks ceding not just revenue but long-term positioning as AI becomes the central capability governments want from their cloud providers. With the DOD actively seeking to modernize with AI, being absent from that procurement cycle would be an increasingly costly choice.

Anthropic and OpenAI have also been expanding government-facing offerings, including classified and sensitive-use programs through government-dedicated cloud environments. Google's move, if completed, would reinforce that all major AI labs are now actively pursuing government defense customers.

The Bottom Line

Google deploying Gemini in classified DOD settings would close a chapter that began with Project Maven. The calculus has shifted: competitive pressure, a more defense-friendly corporate culture, and the sheer scale of government AI procurement have made full abstention from military AI increasingly difficult to justify. If the deal proceeds, it signals that the "no weapons AI" era of Google's AI policy is effectively over.

Related Articles

Sources