Alphabet's Google is in discussions with the U.S. Department of Defense over a potential classified artificial intelligence agreement that would allow its Gemini models to be used inside secure military systems, Reuters said, citing reporting from The Information and people familiar with the matter.
The report claimed that the framework under discussion could give U.S. defense agencies access to Google's AI systems for "all lawful uses" within classified environments. The agreement, however, has not been finalized and has not been publicly confirmed by either Google or the Pentagon.
According to Reuters, Google is simultaneously pushing for guardrails on how its technology would be deployed. These include limits intended to prevent the use of its AI for domestic mass surveillance programs and restrictions on the development or deployment of autonomous weapons systems without human oversight.
The negotiations come as the Pentagon increases its reliance on commercial AI providers to modernize operations and accelerate data-driven decision-making. In earlier reporting, Reuters said U.S. defense officials have been urging leading AI companies, including OpenAI and Anthropic, to expand access to their systems across both unclassified and classified networks. That push is part of a broader effort within the Department of Defense to integrate generative AI into intelligence analysis, logistics planning and administrative workflows.
The same reporting noted that military officials have been seeking multiple vendors rather than relying on a single provider, reflecting concerns about resilience, competition and technological dependence. This multi-vendor approach has opened the door for companies such as Google, OpenAI and Anthropic to compete more directly for sensitive government contracts.
The broader AI contracting environment has also been shaped by shifting relationships between major developers and U.S. defense agencies. Reuters reported that OpenAI has expanded its government-facing offerings by making its models available to U.S. agencies through commercial cloud infrastructure, including Amazon's cloud unit. That arrangement allows federal customers to deploy AI tools within secure environments while relying on commercial hosting frameworks that meet government compliance requirements.
At the same time, tensions between AI companies and defense customers have surfaced over usage restrictions. Anthropic has previously raised objections to certain military applications of its models, particularly around autonomous weapons systems. According to Business Insider, internal discussions across the industry show that companies are increasingly negotiating strict boundaries on how AI can be used in defense contexts, particularly where lethal decision-making or surveillance is involved.
For Google, the reported talks mark a continuation of its gradual return to defense-related work after scaling back direct military involvement several years ago. The company faced internal employee backlash in 2018 over its participation in the Pentagon's Project Maven program, which analyzed drone footage using AI. That episode led Google to publish AI principles that restricted certain military applications, though those policies have since been revisited in the context of commercial and national security partnerships.
According to Business Insider, Google has recently indicated internally that it is increasing engagement with national security customers while maintaining review processes designed to ensure alignment with its updated AI governance standards.
If finalized, the proposed agreement would place Google among a small group of frontier AI developers actively positioned within classified U.S. government environments, alongside competitors already building out defense-related cloud and model deployment infrastructure. The deal would also reflect a broader shift in how the Pentagon is sourcing artificial intelligence capability, moving from experimental pilots toward embedded operational systems.
Neither Google nor the Pentagon has publicly commented on the reported negotiations as of publication.
