Google, historically cautious about military uses of its technology, is negotiating a classified AI contract with the U.S. Department of Defense — a move that could signal a significant policy shift.
The Information reported April 16, citing people familiar with the matter, that Google is in talks with the Pentagon to deploy AI models inside classified environments.
If the talks succeed, Google’s Gemini model would represent a major turning point, extending the company's technology into military and national-security operations.
The discussions center on expanding AI use beyond unclassified settings into classified systems. Last November, the two sides signed an agreement limiting AI use to “legal purposes,” but that deal remained narrowly scoped.
A new agreement could open the door to using AI in sensitive work such as military analysis, intelligence processing and operational support.
Google reportedly sought several guardrails in talks: banning large-scale domestic surveillance, restricting fully autonomous lethal weapons and requiring meaningful human control over AI operations.
Those proposed limits are similar to conditions OpenAI negotiated with the Defense Department. But legal experts warn that the phrase “legal purposes” may significantly dilute the practical effect of such restrictions.
The negotiations mark a sharp departure from Google’s earlier posture. In 2018, after employee protests, Google withdrew from Project Maven, a drone-video analysis program, and stepped back from military partnerships. The company adopted ethical principles opposing the weaponization of AI and limiting surveillance technology — restrictions it reportedly removed after 2025.
Google has been intensifying its push into public-sector and defense work. It has expanded cloud-focused public-sector teams and hired heavily from government and military ranks. The company has set a contract target of 6 billion USD (about 8.8 trillion KRW) for 2025–2027, with defense and national security central to that plan and AI-based deals seen as a key growth driver.
The talks follow a failed negotiation between the Pentagon and rival Anthropic, which insisted on bans on autonomous lethal weapons and large-scale surveillance. Those demands led to a breakdown in talks; Anthropic was later flagged as a supply-chain risk and effectively sidelined from defense contracts. By contrast, Google has pursued a more flexible negotiating strategy to expand its presence in the defense market.
Discussions go beyond simple model access: they reportedly include building AI infrastructure inside classified enclaves, expanding GPU capacity, running Google’s TPU chips in classified environments for the first time, and extending Google Distributed Cloud. The goal is to enable the military to run large-scale AI computations inside its own secure systems.
Internal pushback persists. In March, a group of Google employees circulated an open letter calling for bans on surveillance AI and autonomous weapons, and Chief Scientist Jeff Dean and others publicly backed Anthropic’s position.
Chan Park cpark@aitimes.com
