Google AI staff quit over the company's expanding military contracts
Google DeepMind researchers are resigning over fears that AI used in military targeting isn't getting the human oversight the company promises.
What happened: At least two Google DeepMind employees, research engineer Sofia Liguori and AI researcher Andreas Kirsch, have resigned this month over the company's deepening military contracts with the US and Israeli governments. Liguori told the BBC she felt 'ashamed' watching Google push AI further into military operations, despite a contract that technically bars her from criticizing the company for life. Kirsch said he doesn't believe Google's internal oversight is strong enough to safely deploy AI in military or classified settings. Other current and former staff, speaking anonymously for fear of retaliation, described similar unease spreading inside the company, though some feel powerless to change course.
Why it matters: Google isn't alone: OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia and startup Reflection have all signed similar military contracts this year, each promising AI help with intelligence, logistics and battlefield operations. A former head of OpenAI's geopolitics team warns AI works faster than people can supervise, compressing the time operators have to catch mistakes in high-stakes situations like target selection. The former director of the Pentagon's Project Maven program says the current approach favors speed over rigorous testing - a tradeoff few outside the Pentagon get to weigh in on until something goes wrong.
How it works, plainly: The Pentagon's main AI tool here is Maven Smart System, built by defense firm Palantir using models from several AI companies, including Google. It processes intelligence and sensor data to help personnel identify and select targets, work that used to take analysts far longer by hand. Google's new Pentagon contract, signed this year, lets its AI be used for classified work under 'any lawful government purpose,' but says it shouldn't support domestic mass surveillance or autonomous weapons picking targets without human oversight. The dispute is whether that oversight is real in practice or just contract language.
The wider pattern: This isn't new territory for Google: a 2024 contract with Amazon worth $1.2bn, called Project Nimbus, supplying Israel's government and military with AI and cloud services, triggered earlier staff backlash, though Google said it excluded classified or weapons-related work. AI-targeting systems used by Israel in Gaza were also the focus of a recent documentary alleging high civilian casualties, claims an Israeli minister called false and distorted. Google says every contract is reviewed against its AI principles and employees can raise concerns without retaliation. Several workers told the BBC the real culture is fear of speaking up.