Google DeepMind staff resign over new Pentagon military AI deal

Internal pushback at Google

Employees at Google DeepMind are resigning in protest over the company's increasing involvement with military operations. A London-based research engineer told the BBC that the shift toward military AI made them feel ashamed to work at the firm. They expressed a loss of faith in the company's ability to remain true to its stated ethical values.

The departures follow a 2025 agreement with the Pentagon that permits Google's AI models to be used in classified military environments. While the tools are intended for lawful government purposes, the move has triggered significant internal criticism. Some researchers believe the company's governance is not sufficient to ensure safe deployment in these high-stakes settings.

This is not the first time Google has faced internal opposition regarding military contracts. In 2024, staff also criticized Project Nimbus, a $1.2 billion joint cloud and AI contract with Amazon for the Israeli government and military. These recurring tensions suggest a deepening rift between the company's public image and its defense-related business.

The Pentagon's new AI terms

The new agreement with the Pentagon includes specific restrictions on how Google's AI models can be utilized. The tools may not be used for domestic mass surveillance or for autonomous weapons that select targets without appropriate human oversight. These guardrails are intended to align the technology with international law and human rights.

Google maintains that it uses a consistent process to review all public sector contracts. The company evaluates these deals under its existing acceptable use policies and AI principles. They also provide dedicated channels for employees to report concerns regarding ethics or misconduct without fear of retaliation.

Despite these stated protections, critics argue that the effectiveness of these safeguards is uncertain. There are concerns about whether these rules will stay meaningful once the AI tools are fully embedded in classified military systems. The core issue remains whether the company can prevent misuse in environments where oversight might be limited.

The risks of rapid deployment

A major concern among researchers is the speed at which AI is being integrated into military systems. Retired Lieutenant General Jack Shanahan noted that the philosophy behind these efforts is to deploy technology as fast as possible. He warned that this priority can come at the expense of the rigor and discipline required for proper test and evaluation.

Sarah Shoker, a former head of OpenAI's Geopolitics team, warned that AI can compress the time available for making decisions in warfare. She noted that when hundreds of targets are selected by AI in a short window, it becomes nearly impossible for a human to review the outputs for errors. This creates a risk of overreliance on automated systems during critical moments.

The dangers of such overreliance were highlighted by reports of a bombing in Iran in February that killed at least 123 children. The strike was reportedly caused by an outdated 2017 database that identified a school as a military site. Experts warn that current AI models may not be robust enough to handle such complex tasks accurately.

Broader industry trends

Google is not alone in signing agreements to provide AI tools to the military in 2025. Other major tech companies, including Nvidia, Oracle, SpaceX, Microsoft, Amazon, and OpenAI, have signed similar deals. This trend indicates a significant shift toward the commercialization of AI for national security purposes.

The integration of AI into physical infrastructure, such as the energy grid, also presents unique challenges. While supporters argue that AI can help optimize these systems for cleaner energy, the massive computing power required can strain public resources. These systems also face risks of misuse and could potentially drive up energy costs as demand increases.

As the Department of Defense emphasizes that speed defines victory in the AI era, the pressure on tech companies to deliver fast is intensifying. This creates a complex environment where the need for rapid innovation must be balanced against the potential for security risks and ethical lapses. The ongoing debate centers on whether the technology can be deployed safely while maintaining human control.