Google DeepMind staff resign over military AI contracts with Pentagon and Israel

Resignations over military work

Several employees at Google DeepMind have resigned or are on their way out over the company's military contracts, according to a BBC report. The departures signal that unease over Google's defence work has moved from internal complaints to researchers voting with their feet.

Sofia Liguori, a research engineer based in London, resigned this week after three years at the company. She told the BBC that Google's deepening work with the US and Israeli militaries had led her to lose faith in the firm's stated ethical values, and that she felt ashamed to work there. Notably, she spoke out despite employment terms that she says bar her from publicly criticising Google; she told the BBC she decided to talk anyway because, in her view, "talking about facts is not criticising".

Andreas Kirsch, an AI researcher who resigned earlier this month, went further on the substance. In his view, Google's governance and oversight are not strong enough to deploy AI safely and responsibly in military or classified settings, or to safely develop far more powerful systems in future. He said Google appears to have no right to control or veto how the Pentagon uses its models, and only limited visibility into how they are deployed, and he argued that current models are not robust enough for battlefield tasks.

A Google spokesperson told the BBC that the company works with public sector organisations worldwide under a consistent review process covering its acceptable use policies and AI principles, and pointed to internal channels for raising concerns alongside strict anti-retaliation policies.

What the deals allow

Google's contract with the Pentagon, signed earlier this year, permits its AI to be used for classified military work. The agreement states the models can serve "any lawful government purpose" and should not be used for domestic mass surveillance or autonomous weapons, including target selection, without appropriate human oversight and control. The BBC reports that OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia and the start-up Reflection signed comparable agreements covering uses from transport and planning to battlefield operations.

Much of the concern centres on how such systems behave in practice. The Pentagon's main intelligence platform, Maven Smart Systems, was built by Palantir and draws on technology from several leading AI firms to help personnel select and identify strike targets. Retired Lieutenant General Jack Shanahan, who formerly ran Project Maven, told the BBC the guiding philosophy appears to be deploying as fast and widely as possible, potentially at the expense of rigorous testing and evaluation. The Pentagon's counterargument is that speed is decisive: under secretary Emil Michael said earlier this year that "speed defines victory in the AI era".

Sarah Shoker, formerly head of OpenAI's geopolitics team, told the BBC that AI compresses the time an operator has to decide, and that when hundreds of targets are machine-selected in a short window, meaningful human review of the outputs becomes almost impossible. Critics inside and outside Google question whether the human oversight promised in the company's AI principles actually holds at those critical moments.

Google's Israel work adds to the friction. Staff criticised the 2024 Project Nimbus deal, a joint Google-Amazon contract worth $1.2bn (£960m) supplying Israel's government and military with cloud and AI infrastructure; Google said at the time the arrangement was not directed at highly sensitive, classified or weapons-related workloads. Israel's AI targeting capabilities have also drawn scrutiny, including in the documentary Naza, which examines AI-assisted bombardment of Gaza and high civilian casualties; an Israeli government minister has dismissed the film's claims as false and distorted.

A workforce under strain

Employees told the BBC that staff are afraid to speak publicly about AI safety and warfare because their contracts prohibit criticising the company, and that people fear retaliation and career damage. Liguori said that inside Google "you feel you are being watched, you can't speak".

One current DeepMind employee, speaking anonymously, described a possible "changing of the guard": motivated staff who signed up for the company's original mission could leave, while early-career hires fill their positions. Another said many people who felt strongly about the changed ethics stance had already gone, with some staying to push back and others concluding that a company of Google's size will not change course.

Why it matters

The resignations turn an ethics debate into a concrete talent problem for Google, which competes fiercely for a small pool of senior AI researchers and now risks losing some of its most principled ones. The dispute also stress-tests Google's own AI principles, since the company promises human oversight and alignment with international law while its contract language leaves the depth of that oversight open to question. And because OpenAI, Microsoft, Amazon and others have signed similar military agreements, the backlash at DeepMind may be an early signal of workforce tensions set to spread across the AI industry.