Our technology helped the IDF. I want AI to benefit humanity, not facilitate genocide.
At a London laboratory where some of the world's most consequential artificial intelligence is built, workers have done something unprecedented: they have asked, collectively and formally, for a say in what their creations are used for. Driven by a Pentagon deal that revived old fears about AI and warfare, and by a deeper unease about their technology's role in Gaza, DeepMind's employees have formed the first union at a major frontier AI lab. It is a moment that places the oldest tension in industrial history — who controls the work, and to what ends — at the very center of the technology that may define the century.
- Google's reversal of its own pledge never to build AI for weapons broke a fragile trust, and DeepMind workers responded by seeking the one thing management cannot easily ignore: collective legal standing.
- The Pentagon deal, covering seven of the world's leading AI companies, has compressed a slow-burning moral crisis into an urgent question: at what point does building the tool make you responsible for the harm it enables?
- Workers cite not only hypothetical future dangers but a present one — Google's AI tools were made available to the Israeli military during the Gaza conflict, and at least one employee has named that plainly as facilitation of genocide.
- The union's demands are surgical rather than absolute: an independent ethics body with real authority, the legal right to refuse morally objectionable work, and the threat of research strikes if Google refuses to engage.
- Google's track record — firing 50 Project Nimbus protesters in 2024, dropping its own AI ethics pledge — suggests the company will resist, but a $2.2 billion shareholder coalition watching from the outside may complicate a clean dismissal.
On a Tuesday in early May, workers at Google DeepMind's London office formally requested union recognition — the first such move at any major frontier AI laboratory. The decision had been building since April, when staff voted to unionize after Google announced a deal granting the US Pentagon access to its AI systems. At least 1,000 workers would fall under the joint representation of the Communication Workers Union and Unite the Union, if Google agreed to recognize them.
The Pentagon agreement was the immediate trigger, but the grievances ran deeper. Workers pointed to Google's expanded provision of AI tools to the Israeli military from the early days of the Gaza conflict — part of a broader $1.2 billion cloud contract signed with the Israeli government in 2021. One worker said simply: our technology helped the IDF, and I want AI to benefit humanity, not facilitate a genocide. Another raised the specter of AI enabling authoritarianism, citing the Trump administration's conduct as evidence that the Pentagon was not a trustworthy steward of powerful technology.
The history behind the moment mattered. In 2018, Google employees had successfully pushed back against Project Maven, an AI contract for analyzing Pentagon drone footage, and the company had published principles pledging never to design AI for weapons. Last year, Google quietly dropped that pledge. The reversal galvanized the DeepMind workers and prompted more than 600 Google employees to sign an open letter to CEO Sundar Pichai demanding the company refuse classified military AI work.
The union's demands were specific: a commitment against developing technology whose primary purpose is harm, an independent ethics body with genuine authority, and the right for individual workers to refuse morally objectionable assignments. If Google declined, workers said they were prepared to organize protests and conduct research strikes — withholding labor on core AI development while continuing minor maintenance to avoid easy retaliation.
The company's recent conduct offered little reassurance. It had fired 50 employees who protested Project Nimbus in 2024. The Pentagon contract itself contained language limiting autonomous weapons and domestic surveillance, but that language was non-binding, and Google retained no veto over how the government used the technology. A coalition of shareholders holding roughly $2.2 billion in Alphabet stock had separately demanded transparency about AI deployments in conflict zones. Google did not respond to requests for comment. The workers, for their part, had decided that the moment for quiet compliance had passed.
On a Tuesday in early May, workers at Google DeepMind's London laboratory sent a letter to management requesting union recognition. The move was not impulsive. In April, staff had voted to unionize, driven by a specific anxiety: the company had just announced a deal with the Pentagon to make its artificial intelligence systems available to the US military, and the workers wanted a formal voice in what came next.
The unionization effort at DeepMind marks the first time employees at a major frontier AI laboratory have sought collective representation. At least 1,000 workers would be covered if Google recognized the Communication Workers Union and Unite the Union as joint representatives. The timing was deliberate. Workers had watched the Pentagon reach agreements with seven leading AI companies—Google, SpaceX, OpenAI, Nvidia, Reflection, Microsoft, and Amazon Web Services. Notably absent was Anthropic, which has clashed with the defense department over contract terms despite having its technology widely used by the military.
One anonymous DeepMind worker explained the calculus plainly: concerns about AI being weaponized to empower authoritarianism, whether through military force or surveillance systems, foreign or domestic. The worker pointed to the Trump administration's "capricious Iran war" and its public feud with Anthropic as evidence that the Pentagon was not a trustworthy partner. Another colleague voiced a different but equally urgent worry. Google had provided the Israeli military with expanded access to its AI tools from the early stages of the Gaza conflict. In 2021, the company had signed a $1.2 billion cloud-computing contract with the Israeli government alongside Amazon. "Our technology helped the IDF," the second worker said. "I want AI to benefit humanity, not to facilitate a genocide."
These concerns were not new to Google. The company had faced worker protests for years over the militarization of its technology. In 2018, employees had rebelled against Project Maven, a contract to build AI systems for analyzing Pentagon drone footage. Google had ultimately declined to renew that deal in 2019 and published principles pledging not to design AI for weapons. But last year, the company dropped that pledge. The reversal had galvanized the DeepMind workers. More broadly, over 600 Google employees had signed an open letter to CEO Sundar Pichai in the days before the Pentagon announcement, demanding the company refuse to make its AI available for classified military use.
The Pentagon's statement framed the agreements as essential to American military dominance. "These agreements accelerate the transformation toward establishing the United States military as an AI-first fighting force," defense department officials said. Google's contract included language stating that AI systems should not be used for domestic mass surveillance or autonomous weapons without human oversight. But that language carried no legal weight. The agreement also stipulated that Google had no authority to control or veto what the government did with the technology, provided it was deemed lawful.
The unionization effort came with specific demands. Workers wanted Google to commit not to develop technology whose primary purpose is to cause harm. They wanted an independent ethics body with real authority. And they wanted the right to refuse, on moral grounds, to work on particular projects. If management rejected these demands, workers said they were prepared to escalate—organizing protests and conducting "research strikes," in which staff would abstain from work that significantly improves core products like Gemini, Google's AI chatbot, while continuing minor maintenance to avoid easy detection.
Google's history suggested the company might resist. In 2024, it had fired 50 workers who protested Project Nimbus, the Israeli government contract. At Microsoft, which had supplied cloud storage to the Israeli military for mass surveillance of Palestinians, workers had occupied a company campus with signs reading "No Labor for Genocide." Microsoft had terminated the Israeli military's access only after media reporting forced the issue. Investors, too, were watching. A coalition of shareholders holding roughly $2.2 billion in Alphabet stock had written demanding a meeting and greater transparency about Google's AI deployments in conflict zones and security-sensitive contexts.
The DeepMind workers were not asking for the company to abandon military work entirely. They were asking for guardrails, oversight, and the ability to say no. Whether Google would listen remained unclear. The company did not respond to requests for comment. What was clear was that the workers had decided the moment for quiet compliance had passed.
Notable Quotes
I have joined the union due to concerns about AI being used to empower authoritarianism, whether through military or surveillance applications, both foreign and domestic.— Anonymous DeepMind worker
These agreements accelerate the transformation toward establishing the United States military as an AI-first fighting force and will strengthen our warfighters' ability to maintain decision superiority across all domains of warfare.— Pentagon defense department officials
The Hearth Conversation Another angle on the story
Why did these workers unionize now, specifically? The Pentagon has been interested in AI for years.
The timing matters. Google had dropped its pledge not to weaponize AI last year, and then announced this Pentagon deal. Workers felt the company was crossing a line it had previously drawn for itself. That reversal was the catalyst.
But Google's contract says the AI shouldn't be used for autonomous weapons or domestic surveillance without human oversight. Doesn't that address their concerns?
Only on paper. The language is non-binding, and the agreement explicitly says Google has no right to veto what the government does with the technology, as long as it's deemed lawful. So the guardrails are more like suggestions.
The workers also mentioned Gaza and the Israeli military. How does that connect to the Pentagon deal?
It's part of a larger pattern. Google gave the Israeli military expanded access to its AI tools during the Gaza conflict, and signed a $1.2 billion contract with the Israeli government. Workers feel complicit in that. The Pentagon deal feels like the same logic expanding—their technology enabling military operations they morally oppose.
What happens if Google refuses to meet their demands?
They're talking about research strikes—refusing to work on major product improvements while doing minor maintenance work to stay under the radar. It's a way to apply pressure without being easily detected or fired for insubordination.
Has this worked at other tech companies?
Partially. Microsoft only ended Israeli military access to its surveillance technology after media reporting and worker pressure. But Google fired 50 workers who protested the Israeli contract in 2024. So there's real risk here.