‘Ashamed to work at Google’: DeepMind researchers quit over US military AI contracts

Growing anger over Google's military AI contracts has led several employees at its DeepMind research division to resign, while others are getting ready to leave. They worry that Google's technology could be used in war without strong enough safety measures or meaningful human control.

The departures show growing disagreements inside one of the world's leading AI developers as Google increases its work with the U.S. military. This includes an agreement that allows its technology to be used in secret military operations.

An investigation published by the BBC on Wednesday found that current and former DeepMind employees were worried about the company's military partnerships, how reliable AI systems are in combat, and the possible career damage people could face if they spoke out against Google's decisions.

One person who has left is Sofia Liguori, a research engineer based in London. She resigned this week after working at Google for three years. She told the BBC that she felt ashamed of the company's growing involvement in developing AI for the military.

Liguori said Google's direction had weakened her trust in the company's public promises about ethical AI. She also described a workplace where employees were afraid to raise concerns because they worried about punishment or harm to their careers.

Another employee who left, AI researcher Andreas Kirsch, resigned earlier this month. He told the BBC that Google's internal rules and checks were not strong enough to make sure AI was used responsibly in secret military operations.

Kirsch also questioned whether Google had enough control over what happened to its technology after handing it over to the Pentagon. He said the company had little information about how the military used it and might have little power to stop uses it considered wrong.

Both researchers raised concerns about whether AI systems are reliable enough for military operations. Wrong information or computer-generated suggestions could have life-or-death consequences in these situations. Kirsch argued that today's AI models were not reliable enough and had not been trained well enough for such difficult tasks.

The controversy comes as the Trump administration speeds up the use of AI across the armed forces. Defense Secretary Pete Hegseth has made the technology a major priority and promised to turn the U.S. military into an "AI-first warfighting force".

Google is one of several technology companies that have signed major deals with the Pentagon this year. Others include OpenAI, Amazon, Microsoft, SpaceX, Oracle and Nvidia. Their technologies are being used for military tasks ranging from managing supplies and planning operations to activities on the battlefield.

The agreement Google signed with the Pentagon earlier this year allows its AI models to be used for "any lawful government purpose", including secret military work. It also sets limits on using the technology for mass surveillance of people inside the United States and for autonomous weapons that choose targets without proper human oversight and control.

Google says its contracts follow its AI principles, including promises to keep humans involved and respect international law and human rights. Critics question whether these protections can really be enforced during secret military operations, especially when government agencies control the technology instead of the companies that developed it.

One example of the military's growing use of AI is Maven Smart System, a Pentagon platform developed by the U.S. defense technology company Palantir. The system sorts and processes intelligence information. It uses AI to help military staff identify possible targets and make decisions more quickly.

Palantir has added technology from several AI companies to Maven, allowing it to analyze intelligence and targeting information faster. But the system's growing abilities have also raised concerns about whether human operators have enough time to check the computer's suggestions before acting on them.

Retired U.S. Air Force Lieutenant General Jack Shanahan, who previously led Project Maven, told the BBC that the Pentagon seemed increasingly focused on using AI as quickly and widely as possible. He warned that this could come at the cost of careful testing and evaluation.

Emil Michael, an undersecretary at the U.S. Defense Department, defended the push to use AI quickly. He argued that success in the military increasingly depends on speed in the age of AI.

Google's military-related deals with Israel have also caused controversy. In 2021, Google and Amazon were chosen for Project Nimbus, a cloud computing project worth about $1.2 billion. It was designed to provide advanced digital services to the Israeli government and public sector, including defense agencies.

The project drew strong criticism from Google employees in 2024. Those who opposed it questioned how the technology might be used by the military and whether Google could properly oversee its use.

Google has repeatedly said that Project Nimbus is not meant for highly sensitive or secret military work involving weapons or intelligence services. Even so, the deal has continued to face opposition inside the company, especially as military uses of AI have become more advanced.

The dispute also raises a wider question about the difference between providing computing systems to government agencies and directly helping military systems make decisions. Employees argue that companies have a harder time overseeing how their technology is used when advanced AI tools operate in secret military settings.

Beyond questions about ethics and technology, the BBC investigation also found concerns about Google's workplace culture and limits on employees who want to criticize the company in public.

Liguori said her employment agreement limited her ability to criticize Google even after she left the company. She decided to speak publicly anyway because she believed employees and the public had a right to understand Google's military work.

Speaking about the atmosphere around people who disagree with the company, she told the BBC: "You feel you are being watched, you can't speak."

Other DeepMind employees who spoke to the BBC without giving their names raised similar concerns. One said many colleagues who disagreed with the company's changing ethical direction had already resigned. Others believed they had little chance of changing decisions made by such a large company.

Another employee suggested that researchers who joined DeepMind because of its original focus on science might gradually be replaced by people who were more comfortable with the company's growing military commitments. This could change the organization's culture.

Google rejected claims that its government contracts lacked proper safety measures or that employees could not safely raise concerns.

A company spokesperson told the BBC that Google uses the same procedures to review public-sector contracts around the world. The company checks that its contracts follow its rules for acceptable use and its AI principles. It also provides special channels where employees can raise ethical concerns or report wrongdoing.

The spokesperson added that Google has workplace rules designed to protect employees from punishment for speaking up.

The disagreement highlights a growing challenge for the AI industry: As its most advanced technology becomes part of military operations, how much control can the companies that develop it keep over how it is used? And what happens when the researchers who built the technology no longer trust the safety measures?