Two named departures, one argument
Staff protests at AI labs are usually anonymous. This one has names. Sofia Liguori, a London-based research engineer who spent three years at Google DeepMind, resigned this week and spoke to the BBC even though, she says, her contract restricts her from criticising the company. "Realising my company is pushing more and more for AI systems in military operations makes me feel ashamed to work at Google," she said, speaking in a personal capacity.
After three years at the company, she told the BBC, she decided to resign over Google's increasing work with the US and Israeli militaries: "I don't agree with this direction and this has led me to lose faith in my company being true to its stated ethical values."
Andreas Kirsch, an AI researcher, also resigned earlier this month. He told the BBC he believes "Google's governance and oversight are not sufficient or meaningful enough to deploy AI safely and responsibly in military and classified settings" or "to develop artificial superintelligence safely in the future." Then the sentence the whole dispute turns on: "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."
Note both of Kirsch's verbs: seems. That is a departing researcher's assessment of his former employer's position, not a verified reading of the contract, which is not public in the BBC's reporting. But it is a specific, checkable kind of claim. Either the vendor has audit and intervention rights in classified deployments, or it does not, and Google's response does not say.
The BBC identified two former DeepMind researchers who spoke publicly about leaving over military-AI concerns, and reported speaking with other employees who had resigned or were preparing to leave. It did not establish a total number of departures attributable to the contracts. Another anonymous staff member told the BBC: "A lot of people have left the company that feel strongly about this [how Google DeepMind has changed its ethics position]." That is an impression from inside, not a count, and it is treated here as one.
A separate departure belongs in a different column. In September, DeepMind engineer Robert O'Callahan resigned because, as he put it in his own announcement, his team's work on chip-design tools would make AI "much faster and cheaper", and he believes AI is "already progressing too fast". Different concern, different month. He is not part of the military-contract story, and counting him in would inflate it.
Read the permission, then read the condition
Google's contract with the Pentagon was signed earlier this year and allows its AI technology to be used for classified military work. As the BBC reports the agreement, Google's AI models can be used for "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".
Read those two halves at their exact strength. "Any lawful government purpose" is broad, but it is not unrestricted use: applicable law, other contract terms and exclusions the BBC did not reproduce may still constrain it, and the full contract is not public. The second half is the safeguard the resigning staff are pointing at. Its operative words — appropriate, oversight, control — are not defined in the reporting. Appropriate by whose standard. Oversight performed by whom, with access to what. Control meaning a review step before use, or a power to stop a use already under way.
Google's AI Principles, which staff cite against the company, include implementing AI with "human oversight" and aligning the technology with "international law and human rights". Those principles are Google's own policy commitments. They are not, by themselves, contractual obligations imposed on the Pentagon as a customer. The BBC reported that former and current staff, and experts it spoke to, were concerned about the level of human oversight at critical moments.
That concern has a practical edge. Sarah Shoker, formerly head of OpenAI's geopolitics team, told the BBC that AI "compresses the time an operator has to make a decision", and that "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." Oversight that cannot keep up with the system it oversees is the scenario the clause has to answer. Kirsch added a readiness objection of his own: "Current AI models are not robust enough, and I don't think they have been trained for such tasks either."
What Google said, and what it did not reach
Google's response deserves to be read in full before it is criticised. A 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." And: "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."
Google's reported response emphasises contract review, policy compliance and channels for employees to raise concerns. The material reviewed does not establish whether Google also has operational visibility, audit rights or enforcement mechanisms for the Pentagon deployments at issue. That is the unanswered question, stated precisely. It is not evidence that no such mechanisms exist, and it is not an answer showing that they do.
The wider setting is not in dispute. Google's agreements with Israel have drawn internal criticism before. In 2024 staff criticised Project Nimbus, a joint contract with Amazon worth $1.2bn (£960m) providing Israel's government and military with AI technology and cloud infrastructure; Google said at the time that Nimbus "is not directed at highly sensitive, classified, or military workloads relevant to weapons or intelligence services". And the Pentagon is not hiding its priorities. US Defence Secretary Pete Hegseth has made military AI integration a key priority, saying the Trump administration will become an "AI-first warfighting force", and an under secretary in the US Department of Defense, Emil Michael, said in a statement earlier this year: "Speed defines victory in the AI era." Speed is the environment the oversight clause has to work in.
The dissent channel is part of the dispute
The contracts themselves are now part of the argument. Employees told the BBC that staff are scared to speak out about AI safety and its uses in warfare because their employment contracts prohibit them from publicly criticising the company.
Liguori put her own position in stronger words: "In principle I'm not allowed to criticise Google for the duration of my life." She said she spoke anyway "because I believe it's important for these things to be known and I also think that talking about facts is not criticising". Liguori told the BBC that she understood her employment agreement to restrict public criticism of Google indefinitely. The report does not establish that a court has interpreted or enforced the provision, and Google was not quoted confirming its scope. She also described the atmosphere for colleagues who stay: "people are scared of retaliation and putting their career at risk. You feel you are being watched, you can't speak."
That sits awkwardly beside the company's anti-retaliation assurance, and it is the tension the BBC's reporting actually documents: the channel exists, and the people who left say speaking through it — or past it — carried a personal risk they had to accept. Another anonymous employee was blunter about the internal mood: "Some people are putting up a fight, others are thinking: 'it's a big company. What do you expect to change?'"
The same question, seven more times
The BBC also named seven other technology companies involved in Pentagon AI contracting: similar contracts were signed by OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and the start-up Reflection, to allow AI technology to be deployed across a range of military settings such as transport, planning and battlefield operations. The report does not establish that their agreements contain the same use permissions or oversight language as Google's.
The broader procurement programme raises similar questions for other suppliers, but each company's contractual safeguards and deployment arrangements require separate examination. What can be said today is narrower: at Google, two named researchers have put the enforcement question on the record, and the public material does not answer it.
Sources
- [1] BBC News, "Google experiences resignation backlash over AI military contracts", Christopher Giles, BBC Eye investigations, 21 October 2026 (primary source; full text read first-hand for this article via a complete verbatim syndication of the BBC report, cross-checked against a second syndication; source for the resignations, all quotations from Liguori, Kirsch and anonymous staff, the BBC's description of the Pentagon agreement, Google's response, the Project Nimbus context and the seven other companies named)Read source
- [2] MadRobot, "Google DeepMind Engineer Quits: AI Is Moving Too Fast", 25 September 2026 (secondary source for the separate September resignation of Robert O'Callahan and his stated reason, drawn from his own announcement)Read source