Sofia Liguori spent three years as a research engineer at Google DeepMind's London office. Last week she quit — and didn't mince words about why: the technology she helped build, she said, "automatically makes it easier to escalate the brutality of war."
Two Google deals are at the center of this. The first is a Pentagon contract signed earlier this year, opening the company's models to classified US military work. On paper, it bars autonomous weapons and mass surveillance without human oversight. The second is Project Nimbus, a cloud-and-AI deal with the Israeli government and military that rights groups have criticized since it was signed years ago.
Liguori isn't alone. Earlier this month, researcher Andreas Kirsch also resigned, telling the BBC that Google's internal governance simply isn't adequate for deploying AI responsibly in classified military settings. He said employment contracts effectively bar staff from criticizing the company publicly, leaving many afraid to raise safety concerns at all.
Inside DeepMind, the pushback runs far wider than two high-profile departures. Nearly 200 employees have signed an internal letter demanding the company end its military contracts. Liguori said her attempts to raise the issue internally went nowhere — posts containing the word "genocide" were deleted, managers and HR criticized her tone as too aggressive, and complaints drew only semi-automated replies.
Google declined to address the substance, with a spokesperson saying the company follows a "consistent process" for public-sector contracts, adheres to its AI Principles, and gives employees channels to raise ethical concerns without retaliation. The episode says something broader about the industry: AI labs talk a lot about safety commitments, but military contracts bring in money too large to walk away from — even when a company's own researchers are the ones objecting.
