Google DeepMind: Ethical Tensions and Military Pressures

A Departure Motivated by Ethics

In January, tragic events involving agents from the Department of Homeland Security (DHS) resulted in the deaths of two individuals. In each case, a federal agent used their weapon against an unarmed citizen, leading to fatal consequences.

In light of these incidents, I discovered that Google was providing Cloud services to DHS agencies, which deeply shocked me. I found it unacceptable for federal agents to act in this manner, and I sought to influence my company to cease collaboration with these agencies. My initiative quickly expanded to include a campaign aimed at preventing Google from engaging in unethical military agreements, as the Pentagon increased pressure on AI providers for unrestricted collaborations involving controversial technologies.

My goal was to ensure that Google DeepMind's (GDM) ethical commitments could withstand this pressure. I particularly wanted GDM to maintain its stance against supporting killer robots. I reached out to many influential figures known for their commitment to AI ethics and safety. However, almost all declined to get involved.

Take, for example, Stuart Russell, a renowned AI researcher who has been advocating against autonomous weapons for over a decade. I worked in his lab for several years. At a conference, he publicly agreed to push his organization to support AI providers against government pressure and promised to consult his members. However, neither the statement nor the survey ever materialized.

Another example is Jeff Dean, Chief Scientist at Google and co-leader of the Gemini AI project. In 2018, Jeff had signed a commitment never to support the development or use of killer robots. I convinced him to co-sign an amicus brief in support of Anthropic against the Pentagon. However, I also asked him to use his influence to prevent Google from entering into an unethical military agreement, but I doubt he did. He remained at Google despite his commitment.

I drafted a 25-page proposal, including contractual clauses and oversight mechanisms. Experts in military law and oversight praised this proposal, which represented an ethical alternative that Google could have adopted. I forwarded this proposal to Demis Hassabis, CEO of GDM, who passed it on to senior political staff. However, it went unanswered until Google signed an agreement.

Google's management claimed they would not sign an agreement. I disagreed, but my warnings were largely ignored. Although I managed to sow doubt within the Pentagon regarding the agreement, Google ultimately signed a contract transferring their AI without restrictions against killer robots or mass AI surveillance. Google's contractual restrictions were even weaker than those of OpenAI. At this point, I could no longer stay at Google in good conscience, so I decided to leave.

This essay recounts why I left Google DeepMind. It is also a broader story: how powerful individuals and institutions failed to uphold their ethical promises in the face of pressure.

Google and the Immigration Supply Chain

After the death of Alex Pretti, I was determined to act effectively. To understand how to mitigate the harm caused by the DHS, I examined the involvement of major tech companies. While Microsoft and Amazon had more significant commitments, I was surprised to discover Google's contracts with the DHS:

• The DHS 2025 AI use case inventory lists Google among the GenAI providers used to "improve operational efficiency" of the DHS.

• Google sells Cloud services to ICE through third parties like ITC Federal.

• On October 3, 2025, Google removed applications that alerted users to ICE activities.

• Google voluntarily handed over the account of a protesting student to ICE without notice, violating their promise of "notifying the user by email before disclosing information [to the government]."

In light of these findings, I wondered how to act effectively. Typical activist actions, such as petitions, had already failed. Google had ignored a large petition on this issue. Moreover, Google’s leadership had likely fortified their company against traditional organizing tactics. Sit-ins, strikes, even a mass resignation of Google engineers: I considered all of them ineffective (if I could even pull them off).

Upon reflection, I concluded that Google would not care about 100 random research engineers resigning. In the AI industry, talent is concentrated, and teams are led by a few stars who are hard to replace. I didn’t need to coordinate 100 engineers. Maybe I just needed to coordinate 10.

I had followed the news and guessed that Sundar Pichai (CEO of Google) was more of a businessman than someone who would say, "give me a grand speech about ethics and I will change my mind." But if a few irreplaceable people were willing to leave, it would matter to the company, so Sundar might listen.

That’s when I remembered seeing Jeff Dean tweet about ICE's bad reputation, retweeting quotes from Anne Frank. Perhaps I didn’t even need 10 engineers; I just needed one.

Discussion with Jeff Dean

At first, I thought about who could connect me with him. But (and this is a good general lesson) if you want to talk to someone about something, you can always JUST ASK THEM!

I told Jeff that I respected him for speaking out, that I wanted Google to disengage from the DHS supply chain. I asked him if he shared these goals and, if so, how I could help.

He suggested it would be reasonable for me to contact a few people. Their names: Sundar Pichai (CEO of Google), Demis Hassabis (CEO of Google DeepMind), and Thomas Kurian (CEO of Google Cloud). I thought, "sure, Jeff. No problem. I’ll just tell them what I think."

My Email

I am writing as a concerned employee at GDM. I recently reached out to Jeff Dean regarding my concerns. He suggested it would be reasonable to send an email to all three of you.

I have no issue with Google working with legal administrations from either political party in the U.S. My concern is not about politics, but rather the events facilitated by Google’s role in the DHS supply chain.

I believe that ICE has largely exceeded its legal mandate to remove illegal immigrants from the country in an orderly manner. According to observers, ICE operations frequently deprive targets of their right to a trial. These operations regularly detain citizens in facilities operating with little (or no) legal oversight. Over 1,000 people are missing in such places. In other locations, the ACLU reports human rights abuses and a severe lack of safety for detainees.

These are not standard and legitimate law enforcement activities. These operations are concerning from a human rights perspective and also pose a reputational risk for any provider involved.

On January 28, 2026, the DHS released its 2025 AI use case inventory, which lists Google as one of several GenAI providers that "improve operational efficiency" of the DHS. I urge Google to immediately cease working with ICE (and DHS more broadly), including by supporting third-party integrators facilitating these specific operations. Whether through a direct ("prime") contract or via third parties (like ITC Federal), Cloud and Gemini must not fuel these operations.

History will judge the tech sector by its involvement in these events. I love working at Google, and I want to ensure that Google is on the right side of this story.

Alexander Matt Turner
Researcher, Google DeepMind

They never responded. I went back to Jeff and asked for lunch to discuss constructive opportunities for real change within Google. I told him, "anytime, anywhere. I’ll come to Mountain View to meet you."

At this point, I thought this was where "Plan A" would fail. To my surprise, he actually agreed, for lunch a few weeks later.

A lot was going to happen by then.

The Pentagon Attempts to Intimidate Anthropic

On February 25, 2026, the Pentagon pressured the cutting-edge AI lab Anthropic to remove the restrictions from its existing contract. These restrictions concerned lethal autonomous weapon systems and AI surveillance/profiling. The ultimatum was clear: "give us your product or we will designate you as a risk to the supply chain." The government wanted to use AI for "any legal use."