DeepMind staff ask a UK tribunal to force a union vote over Google’s military AI work

WIRED reports that Google DeepMind employees in the UK have gone over management’s heads. The Communication Workers Union and Unite the Union, which staff put forward as their representatives after voluntary-recognition meetings stalled, filed with the Central Arbitration Committee, the government body that settles employer-union disputes. They are asking it to confirm there is enough support to trigger a recognition ballot. If it succeeds, roughly 1,000 staff in product-focused roles would qualify, and the company would be legally bound by the result. Organizers emailed all UK staff urging them to take part in a ballot expected in the coming months.

It would be among the first serious attempts to unionize a frontier AI lab, and organizers want it to be a playbook for OpenAI and Anthropic staff in the UK.

According to WIRED, the push began in February 2025 when Alphabet removed a pledge from its AI principles saying it would not pursue uses like weapons development. Organizers object specifically to agreements with the U.S. and Israeli militaries and say they would demand Google back out of those contracts if they win.

It is healthy for the people building these systems to want a formal say in how they are used. One anonymous employee told WIRED there are few checks, from employees or regulators, on an industry with this much economic power. Internal dissent at AI labs has mostly meant open letters and resignations, as at OpenAI (a safety employee’s resignation). A union with collective bargaining is a sturdier instrument.

A union represents its members, and they won’t all agree on defense work. Google says it is proud to support national security, while committing not to use AI for domestic mass surveillance or autonomous weapons without appropriate human oversight. Many people, including some inside the lab, think democracies should have access to frontier AI; the Pentagon already uses tailored commercial models (coverage). Bargaining over which customers a company may serve is unusual territory, and a recognition win may not translate into leverage over contracts.

CWU’s John Chadfield says talks were handed to HR and treated as box-ticking. DeepMind says it listened to supporters and opponents and acted in good faith, respects employees’ right to decide and will participate constructively. Escalation to a government arbitration body suggests trust has broken down.

If this scales, worker voice becomes a real governance layer inside AI labs alongside boards and regulators, which could be a meaningful check. It could also make labs more cautious about where they base staff. Labor organizing at AI companies is part of a broader question about who gets a say in how the technology is deployed (Anthropic’s study of its own workforce). More in our Google coverage and future of work coverage.