Staff inside Google DeepMind in London are pressing the company to release them from lifetime non-disparagement clauses that, they say, stop them from warning the public about the risks of the artificial intelligence systems they build. The push, described by two current employees who spoke anonymously for fear of retaliation, asks Google to remove the clauses from existing contracts, not only from the terms offered to new hires. Google says it no longer enforces the clauses after employees leave, but the staff argue that promise alone does not give them legal certainty.
What the contracts say
The clauses sit in contracts signed by DeepMind staff in Britain, where London is a major base for Google’s artificial intelligence division. A contract reviewed by the Washington Post forbids some staff from criticizing the company for life, even after they leave. The employees said the agreements block them from speaking publicly about the potential impacts of artificial intelligence.
How the dispute surfaced
The dispute came into the open after former DeepMind engineer Sofia Liguori told the BBC last month that she had ignored her own contract restrictions to speak publicly against Google building military artificial intelligence. Her account prompted current DeepMind workers in Britain to check their own contracts, and some found the same lifetime clauses.
Dozens of employees then joined an internal discussion about the language, with some criticizing it directly and others posting an emoji reading 1984, a reference to the dystopian novel. One employee said the clause still creates fear even if it may not hold up in court, because British whistleblower protections cover disclosures made in the public interest.
What Google says
On Friday, a Google executive told staff in the internal discussion that the company would not enforce the clause for workers who had left. A spokesperson said Google has always encouraged open dialogue about safety and responsibility, and that recent updates to new-hire terms removed the post-termination aspect of the clauses.
The company also said it does not enforce non-disparagement clauses after people leave Google, and that it has dedicated internal channels for workers to escalate concerns to senior leaders. Current staff told the Washington Post that is not enough: they want the clause removed entirely from existing contracts, not only from the terms offered to new hires.
Our read
Google’s partial concession reveals the gap between a promise and a contract. A statement that a clause will not be enforced is reversible in a way that deleting it is not, and researchers weighing whether to speak publicly will read that difference closely.
The dispute also puts Google out of step with its rivals. Staff at Anthropic and OpenAI regularly comment publicly about AI dangers, while fewer Google voices have joined that debate. If DeepMind’s researchers stay quiet, the public conversation about AI risk loses the people who know most about one of its leading systems.
There is also a longer history here. Google once positioned itself as a place where internal disagreement was welcomed, then dropped a Pentagon drone-imagery contract in 2018 after employee protests, and later fired workers who protested its work with the Israeli military in 2024. The non-disparagement fight is the latest chapter in that shift.
Why It Matters
The stakes go beyond one company’s employment contracts. The researchers building frontier AI systems are among the few people who can credibly warn the public when those systems behave dangerously, and their warnings often arrive before regulators or journalists can document a problem.
If their contracts chill that speech, the public learns about risks later, or not at all, and policymakers debate AI safety without testimony from the people closest to the work. That is why a clause that looks like routine employment paperwork has become a test of whether the AI industry’s safety debate includes its own builders.
Background
DeepMind, which Google acquired in 2014, is the company’s London-based artificial intelligence research division and one of the field’s most prominent labs. In September, longtime DeepMind leader Demis Hassabis joined the chief executives of OpenAI and Anthropic and Elon Musk in calling for a slowdown in AI development to allow more research into keeping the technology under human control.
Source: Washington Post report and NDTV report
0 Comments