The key facts
- Research engineer Sophia Liguori and researcher Andreas Kirsch publicly announced their departure from Google DeepMind over the company's military contracts.
- This year, Google and a number of other tech companies signed contracts with the Pentagon to use AI, including the Maven Smart Systems platform.
- Google says it complies with its AI principles and acceptable use policies; employees speak of fear of retaliation for criticism.
Resignations over military contracts
Google is facing discontent among some employees over the use of its AI technologies for military purposes. The BBC spoke with several Google DeepMind workers who recently resigned or are in the process of leaving. Sophia Liguori, a research engineer from London, resigned this week and, despite contractual restrictions, told the BBC about her concerns. 'The realization that my company is increasingly pushing AI systems into military operations makes me ashamed to work at Google,' she said, emphasizing that she was speaking in a personal capacity. Earlier this month, researcher Andreas Kirsch also resigned, believing that 'governance and oversight at Google are insufficient for the safe and responsible deployment of AI in military and classified settings' and that current AI models are 'not reliable enough' for the battlefield.
Pentagon and Maven contexts
Earlier this year, Google and a number of other technology companies — OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia and the startup Reflection — signed contracts with the US military to deploy AI in classified military environments, including transport, planning and combat operations. US Defense Secretary Pete Hegseth called the integration of AI into the armed forces a priority, stating that the country would become an 'AI-first military force.' A key example of military use is Maven Smart Systems, the Pentagon's platform for organizing intelligence, developed by Palantir: it helps personnel select and identify targets for strikes. Retired Lieutenant General Jack Shanahan, former director of Project Maven, believes the philosophy now is to 'deploy as fast and as widely as possible, possibly at the expense of testing rigor.' The US Department of Defense, however, stated that 'speed determines the winner in the AI era.'
Contract terms and oversight concerns
Google's contract with the Pentagon permits the use of its AI models 'for any lawful government purposes' and states that they 'must not be used for domestic mass surveillance or autonomous weapons (including target selection) without appropriate human oversight and control.' Former and current employees and experts expressed concern to the BBC about the level of human oversight at critical moments. Sarah Shoker, who previously led OpenAI's geopolitics team, noted that AI reduces decision-making time, and with hundreds of AI-selected targets, it is 'almost impossible for an operator to meaningfully verify the results for errors.'
Google's position and the atmosphere inside the company
In response to a BBC inquiry, Google stated: 'We work with government organizations around the world and follow a single process for reviewing and concluding such contracts, including compliance with our acceptable use policies and AI principles,' and also noted the existence of channels for employees to raise concerns and strict anti-retaliation policies. At the same time, employees told the BBC that they are afraid to speak openly about AI safety, since the contracts prohibit public criticism of the company. 'In principle, I am forbidden from criticizing Google for the rest of my life,' said Liguori, explaining that her decision to speak publicly was driven by a desire for 'these things to become known.' A current DeepMind employee, who wished to remain anonymous, suggested that the company may be experiencing a 'changing of the guard': motivated employees are leaving, but their places are being taken by junior specialists.
Project Nimbus and Israel
Google's agreements with Israel have also been controversial: in 2024, employees criticized the joint contract with Amazon worth $1.2 billion (£960 million) — Project Nimbus, which provides the Israeli government and military with AI technologies and cloud infrastructure. Google said at the time that the project 'is not directed at highly sensitive, classified or military tasks involving weapons or intelligence services.' In addition, Israeli AI capabilities in the war came under scrutiny after the release of the documentary 'Naza' about the role of AI targeting systems in strikes on the Gaza Strip, which caused high civilian casualties. An Israeli government minister stated that the claims in the film are false and distorted.
Image source, Reuters DeepMind is Google's AI company
Image source, Supplied Sofia Liguori resigned from Google, saying the company's push for more AI technology in the military made her feel
Image source, Getty Images US Defence Secretary Pete Hegseth said in January that the US
Image source, Supplied Former Google employee Andreas Kirsch does not think current AI models are
Image source, Justin Sullivan/Getty Images Google's AI Principles include implementing AI with
