Several Google DeepMind researchers have resigned or are reportedly preparing to leave over the company’s expanding military contracts, according to BBC interviews with current and former staff. The departures follow Reuters reporting that Google signed a classified agreement with the US Department of Defense permitting its AI models to be used for “any lawful government purpose,” including classified military work, raising concerns similar to those prompted by a recent surveillance app built by US operatives to monitor domestic political groups.
The Researchers Who Walked Out
Two named researchers have publicly attributed their departures to concerns about military deployment and the limits of Google’s oversight once its technology enters classified environments.
Sofia Liguori, a London-based research engineer who spent three years at Google DeepMind, told the BBC she resigned after losing confidence in the company’s commitment to its stated ethical values, specifically its growing work with US and Israeli militaries.
AI researcher Andreas Kirsch resigned separately, stating that Google’s governance and oversight structures were insufficient for safe and responsible deployment in military and classified environments, per the BBC. Kirsch also reportedly said Google may have limited practical ability to monitor or veto how the Pentagon uses its technology once those systems are integrated into classified infrastructure, according to the BBC.
What the Contract Says
The reported agreement includes restrictions on certain uses, but also a provision that may allow government-requested changes to some of Google’s safety settings.
The DoD agreement states that the systems should not be used for domestic mass surveillance or autonomous weapons without appropriate human oversight and control, according to Reuters. That provision offers some assurance on paper.
The second clause deserves close attention. Reuters also reported the agreement requires Google to help adjust its AI safety settings and filters at the government’s request, a provision that could allow government-requested changes to some of those safeguards.
Project Maven provides relevant context. The Pentagon initiative, developed with Palantir, uses AI to organize intelligence data and assist personnel with identifying and selecting targets. Pentagon officials have described speed as central to victory in AI-driven military environments, per BBC reporting.
Retired Lt. Gen. Jack Shanahan, the former Project Maven director, warned that deployment pressure can outpace rigorous testing and evaluation, also according to the BBC. Former OpenAI geopolitics head Sarah Shoker described the practical problem to the BBC this way: AI can compress the time available for decisions. That compression makes it difficult for a human operator to meaningfully evaluate hundreds of AI-selected targets for errors.
Google’s Position and the Broader Industry
Google says its contract-review process is consistent and that employees can raise concerns without retaliation, but the company is far from alone in pursuing military AI arrangements.
Google told the BBC it applies a consistent contract-review process aligned with its acceptable-use policies and AI principles, and that employees have channels to raise concerns without fear of retaliation. The company’s published AI principles require human oversight and alignment with international law and human rights.
BBC reporting confirms Google is not operating in isolation: OpenAI, Amazon, Microsoft, Amazon, and Google alongside SpaceX, Oracle, Nvidia, and Reflection have all signed or pursued US military AI arrangements.
Google and Amazon’s $1.2 billion Project Nimbus contract has drawn separate employee criticism over cloud and AI services provided to Israel’s government and military. Google said at the time that Project Nimbus was not intended for sensitive military or intelligence workloads involving weapons or intelligence services. Claims about AI-assisted targeting connected to that contract have been publicly contested. An Israeli government minister rejected the assertions in the documentary Naza as false and distorted, according to the BBC.
The Question These Resignations Leave Open
What remains unresolved is whether “human oversight” will mean substantive, accountable review or simply a human presence in a process moving faster than any individual can reliably track.
That question cannot be answered by contract language alone. For anyone building AI systems intended to reflect ethical commitments, that ambiguity is the whole problem.
