Google-Pentagon: The Internal Revolt Against Military AI

Growing Opposition at Google

On Monday, over 600 Google employees, including researchers from DeepMind and around twenty directors, expressed their disagreement with the company's management. They sent a letter to Sundar Pichai, the CEO of Google, requesting the cessation of negotiations regarding the deployment of Gemini on the Pentagon's classified military networks. The employees are concerned that Google will lose all control over the use of its AI models in these highly secure environments.

The letter emphasizes a refusal to engage in any classified work related to military AI for the Pentagon. In 2018, a similar petition gathered 4,000 signatures against Project Maven, a program for object recognition in drone video feeds. At that time, Google succumbed to pressure and did not renew the contract, allowing Palantir to take over the project. Today, the mobilization is more modest but takes place in a very different context.

Strengthened Ties with the Pentagon

Since the backlash of 2018, Google has gradually restored its relations with the Pentagon. In February 2025, the clause prohibiting the use of AI for weapons and surveillance was removed, justified by Demis Hassabis, CEO of Google DeepMind, as necessary to remain competitive globally. In December 2025, Google launched GenAI.mil, powered by Gemini, for all U.S. defense personnel. In March 2026, Gemini agents were deployed to three million Pentagon personnel, but only at an unclassified level.

Current negotiations aim to expand the use of Gemini to classified networks, with a "all legal uses" clause that the Pentagon deems essential for its flexibility. Google has proposed safeguards against mass surveillance and autonomous weapons, but the letter's signatories believe these protections are unenforceable, highlighting that the Pentagon prohibits any imposition of external controls on its AI systems.

An Ethical and Strategic Debate

On air-gapped networks, physically isolated from the internet, Google would have no access to queries or decisions made. Sofia Liguori, an AI research engineer at Google DeepMind and a signatory of the letter, expressed to Bloomberg her concerns about the company's vague response, which asks employees to trust management to negotiate appropriate contracts.

The letter also mentions a report on similar negotiations between the Pentagon and OpenAI, shortly after Anthropic was excluded from Defense supply chains for refusing the "all legal uses" clause. A third of the signatories requested anonymity for fear of retaliation, but among those who agreed to be named are more than twenty directors and senior researchers from DeepMind.

Despite this opposition, Google's management appears determined to continue its commitments with the Pentagon. In 2018, it took just twelve resignations and 4,000 signatures to influence the company's decision. In 2026, 600 names stand against a classified market worth tens of billions of dollars, with a Pentagon ready to sanction companies that do not comply with its demands.