DeepMind Scientist Resigns in Anger, Exposing Google's Sins! Accuses Company of Compromising on Pentagon's Military AI Demands, Ignoring Safety

Former Google (GOOGL-US) DeepMind AI safety researcher Alex Turner recently published a detailed account explaining his resignation in June, directly accusing Google of signing an 'unrestricted' military AI contract with the U.S. Department of Defense (Pentagon) and criticizing several industry heavyweights for remaining silent at a critical moment, failing to prevent the agreement.

Turner stated that the incident began with a case earlier this year in which Department of Homeland Security (DHS) agents shot and killed two civilians on the street.

Upon learning that Google Cloud services were part of the DHS supply chain, Turner felt deeply uneasy and launched an internal campaign urging the termination of the related collaboration. He later expanded his focus to the Pentagon’s ongoing pressure on major AI companies to sign military AI contracts with no restrictions.

Turner believes that signing such a contract would mean these companies’ AI models would be used to develop lethal autonomous weapons or build large-scale AI surveillance networks. Anthropic, for instance, reportedly faced threats of being cut off from supply chains after refusing similar terms.

To persuade company leadership, Turner drafted a 25-page military AI oversight framework, including human monitoring mechanisms, arguing that AI should not be used in lethal weapon systems or mass surveillance without human control.

He said the proposal received high praise from experts in military and surveillance law and garnered support from approximately 250 employees in the DeepMind and Google research divisions through an internal petition.

Nonetheless, Turner pointed out that his communications with Google’s senior management ultimately failed to yield substantive progress. He directly contacted DeepMind CEO Demis Hassabis, who replied that two senior policy officials would evaluate the proposal, but subsequent communication stalled, and the proposal reportedly never received a formal response from top leadership.

Turner mentioned that he had placed hope in Google’s Chief Scientist Jeff Dean, who had publicly co-signed a statement opposing lethal autonomous weapons in 2018.

Turner originally hoped Dean would leverage his influence to pressure the company, even threatening resignation to force management to compromise.

According to Turner, Dean did take some action, including publicly co-signing documents supporting Anthropic’s resistance against Pentagon pressure, which reportedly made the Pentagon cautious about Google’s stance.

However, Dean ultimately did not further use his influence to block the contract signing, leaving Turner disappointed.

Turner further stated that he personally approached Stuart Russell, president of the 'International Association for Safe and Ethical AI' (IASEAI), and Turing Award winner Yoshua Bengio for public support during a conference in Paris hosted by IASEAI.

However, Bengio’s office later replied that they would not issue a statement; Russell had promised during the event to push for a full membership vote and public statement, but Turner noted that this commitment was never fulfilled after the conference, and the association did not respond to his repeated follow-up attempts.

Google’s Contract Signing and Subsequent Controversy

Turner stated that Google formally signed the military AI contract with the Pentagon at the end of April and did not publicly announce the matter.

He specifically pointed out that clauses restricting autonomous weapons and mass surveillance used non-legally binding language.

In response to external criticism, Hassabis stated in an interview that the company’s AI principles 'have not changed.' In fact, the original agreement when Google acquired DeepMind in 2014 explicitly stipulated that AI would not be used for military or weapons purposes, but that clause was removed by Hassabis in the February 2025 revision of the principles.

Turner also raised technical concerns. He noted that since the servers for the relevant AI systems are physically deployed within the military’s independent data centers, Google has effectively lost the ability to monitor the 'chain of thought' operations of the models, and the military may not have engineers with AI alignment training backgrounds on-site for real-time oversight—he believes this poses a safety risk.

At the end of his statement, Turner said he felt disbelief upon learning of the contract signing and described returning to the office as feeling like coming back to a place that no longer belonged to him.

After resigning, Turner did not join another company like some industry figures but chose to temporarily step away from the industry. He said: 'Whenever I feel like backing down, I look at the photo saved on my phone—the civilian who was innocently shot dead by DHS agents on the street.'