Most AI stories on this site are about something an AI system did. This one is different. Nothing went wrong on a screen. Instead, some of the people who build the technology have decided they cannot keep doing it.
On 21 October 2026, BBC Eye reported Google staff resign in backlash over its AI contracts with military. The BBC says it spoke to several tech workers who have recently resigned from Google DeepMind, Google's AI company, or are in the process of leaving.
Who is leaving, and why
The BBC names two people.
Sofia Liguori is a London-based research engineer. The BBC reported that she had resigned that week, after three years at the company. Speaking in a personal capacity, she told the BBC: "Realising my company is pushing more and more for AI systems in military operations makes me feel ashamed to work at Google."
Andreas Kirsch is an AI researcher. The BBC says he resigned earlier this month. He told the BBC that he believes "Google's governance and oversight are not sufficient or meaningful enough to deploy AI safely and responsibly in military and classified settings."
He also told the BBC: "Google seems to have no right to control or veto how the Pentagon uses its AI and also seems to have limited visibility into how it's being deployed."
These are their views. Google has not confirmed the departures by name.
What Google says
A Google spokesperson told the BBC:
"We work with public sector organisations worldwide and follow a consistent process for reviewing and entering these contracts, including compliance with our acceptable use policies and AI principles."
"We also have dedicated channels for employees to safely raise any concerns - ranging from business ethics to misconduct - and have strict workplace policies in place to prevent retaliation of any kind."
What the deal is, as far as we can see
On 1 May 2026, the Pentagon announced Classified Networks AI Agreements with eight frontier AI companies, Google among them. The release says their AI will run on its classified networks for lawful operational use. The US department's own website now calls it the Department of War.
The BBC reported that Google's agreement lets its AI models be used for "any lawful government purpose". That is the BBC's account of the agreement, not an on-the-record quote from anyone. The contract itself is not public. The BBC also reported that the agreement says the models should not be used for domestic mass surveillance or autonomous weapons, including target selection, without appropriate human oversight and control.
The tech workers told the BBC they are particularly worried about this year's deals, because they greatly widen what the Pentagon can use the technology for.
The rules that changed
In June 2018, Google published AI at Google: our principles. It included a short list of AI uses it would not pursue, among them weapons meant to cause injury to people and surveillance that breaks internationally accepted norms.
On 4 February 2025, Google announced updated principles in Responsible AI: Our 2024 report and ongoing work. The 2025 rewrite dropped the 2018 will-not-pursue list. The current AI Principles page talks instead about human oversight, due diligence and feedback, in line with international law and human rights.
To be fair to Google, that is not the same as saying it builds weapons. No source we have read says Google's AI has caused any specific harm in war. What changed is the shape of the promise: a clear list of things Google would not do became broader goals and checks.
The human side
Employees told the BBC that staff are scared to speak out because their employment contracts stop them from publicly criticising the company. Liguori said: "In principle I'm not allowed to criticise Google for the duration of my life." She added: "people are scared of retaliation and putting their career at risk. You feel you are being watched, you can't speak."
A current Google DeepMind staff member, speaking to the BBC without being named, put the mood simply: "Some people are putting up a fight, others are thinking: 'it's a big company. What do you expect to change?'"
Google says it has safe channels and strict policies against retaliation. Both can be true at once: the channels can exist, and staff can still feel afraid to use them.
The AI side
Underneath all this sits a plain technical question. Is today's AI ready for this kind of work?
Kirsch does not think so. "Current AI models are not robust enough, and I don't think they have been trained for such tasks either," he told the BBC.
Sarah Shoker, formerly head of OpenAI's Geopolitics team, told the BBC that speed is part of the risk. "When you have hundreds of targets that are selected by AI and a short amount of time, it is almost impossible for a human operator to meaningfully review the AI's outputs for errors."
The phrase "human oversight" sounds reassuring. But it only works if a person has the time, the information and the power to say no.
The Pentagon sees it differently. "Speed defines victory in the AI era," Emil Michael, an under secretary in the US Department of Defense, said in a statement earlier this year, according to the BBC.
What we do not know
- How many people have left. The BBC says it spoke to several, and gives no total.
- What the full contract says.
- How Google's AI is actually being used on those networks.
What to sit with
- If a company swaps a clear list of things it will not do for broader goals, how would anyone outside know where the line now sits?
- What does "human oversight" really mean when the person checking has only moments to decide?