Google's Gemini AI Connects to the Pentagon: The Security Boundary Dispute Between the Military and Tech Companies

Google recently deployed its Gemini AI agent system to the U.S. Department of Defense's non-classified network environment, aiming to assist with daily tasks and decision support. Although this move does not involve sensitive data, it has sparked an important discussion about the boundaries of AI applications in military scenarios.

The current deployment is strictly limited to non-classified networks, primarily to allow the military to evaluate the stability, reliability, and compatibility of the AI system with existing policies. At this stage, Google has set clear security restrictions for the model, prohibiting certain high-risk uses, while the Department of Defense argues that it should support "all legal uses." There is tension between the two sides regarding the balance of autonomous decision-making and human oversight.

This contradiction is not isolated. Another AI company, Anthropic, has publicly stated that it refuses to use its technology for fully autonomous weapon systems or large-scale domestic surveillance. This stance weakens the feasibility of some military AI programs and prompts the Pentagon to emphasize ethical constraints and mandatory technical safeguards in procurement contracts. Whether to extend the AI system to classified or top-secret networks in the future remains without a public timeline. However, it is foreseeable that the Department of Defense must seek a new contractual framework between operational efficiency and ethical compliance while promoting technological applications. The security red lines of technology companies and the military's pursuit of effectiveness are jointly shaping the policy direction of the next generation of defense AI.