Researchers at Google DeepMind are publicly pushing back against employment contracts they say bar them from criticizing the company about AI risks forever — including after they quit — and Google has now conceded ground, saying it will not enforce the post-employment restrictions and has already removed them from terms for new hires.
The Washington Post reported Tuesday that DeepMind staff are demanding the freedom to tell the public what they know about AI, protesting what the report describes as a lifetime ban on criticizing the company. The dispute has unfolded in internal forums at Google before spilling into public view, and it touches one of the most contested questions in the AI industry: whether employees at the labs building frontier systems can warn the outside world about risks they witness on the job. For more context on this story, see our ongoing artificial intelligence updates.
What the Contracts Say
According to the Post's reporting, DeepMind employees' contracts have included non-disparagement clauses that continue to bind workers even after their employment ends. Under such terms, a departing researcher who publicly criticized Google's safety practices — or simply disclosed information about the company's AI systems — could theoretically face legal exposure for the rest of their career.
A participant in the internal discussion described the arrangement in comments quoted by Hacker News readers discussing the article: the contract apparently included language specifying that quitting does not release an employee from the non-disparagement obligation. For safety researchers, whose professional value often depends on their ability to publish and speak openly, the effect is a form of permanent silencing.
Google's Response
The pressure appears to have produced results. On Friday, a Google executive responded to the internal discussion about the non-disparagement clause and said the company would not enforce it against workers who had left the company, according to the Post.
By Tuesday, a Google spokesperson had gone further, telling the newspaper in an email: "We've always encouraged an open dialogue about safety and responsibility." The spokesperson added: "Recently, we've updated our terms for new hires, which included removing the post-termination aspect of non-disparagement clauses."
That is a meaningful concession. Post-employment non-disparagement clauses have become a flashpoint across the technology industry precisely because they prevent the public from hearing about problems inside companies after employees have already left — the moment when individuals have the least to lose and the most credibility. Google's decision to strip the post-termination component from new-hire contracts suggests the company judged the reputational cost of defending the clauses to be higher than whatever confidentiality benefit they provided.
A Long-Running Tension at the AI Labs
The dispute fits a pattern that has played out across the frontier AI industry. Safety researchers have repeatedly found themselves caught between confidentiality obligations to their employers and a conviction that the public deserves to know about risks posed by systems more capable than anything deployed before.
DeepMind itself has seen prominent departures tied to safety concerns, and rival labs have faced their own high-profile resignations from researchers who said they could no longer remain silent about how their employers were approaching risk. Employee petitions and open letters have become a recurring mechanism through which AI lab staff attempt to influence their companies' conduct — on military contracts, on safety commitments, and on transparency.
The DeepMind protest adds a new legal dimension to that dynamic: rather than disputing what the company is building, employees are disputing the contractual machinery that governs whether they may ever talk about it. Employment lawyers note that the enforceability of broad non-disparagement clauses varies by jurisdiction, and regulators in the United States have previously moved against companies that used overly broad separation agreements to silence departing workers.
The timing also matters. Google is competing for AI research talent against OpenAI, Anthropic, Meta, and a growing cohort of well-funded startups, and the researchers most likely to weigh risk questions are the ones most sensitive to whether they could ever speak publicly about what they see. A company known for gagging former employees faces a harder recruiting pitch to exactly those candidates — a commercial calculation that may have weighed on Google's decision to update its terms so quickly.
Why It Matters
Frontier AI companies occupy an unusual position. They employ some of the few people on Earth with hands-on knowledge of the most capable AI systems, yet those employees are typically bound by dense confidentiality agreements that outlast their employment. If every credible witness to a lab's internal safety practices is legally constrained from speaking, the public record depends entirely on what companies choose to disclose about themselves.
That is why the outcome of this dispute extends beyond Google. How AI companies constrain — or protect — their researchers' future speech may shape whether warnings about powerful systems reach regulators, journalists, and the public early enough to matter, or arrive only in retrospect. Google's concession removes one lifetime gag clause from its new-hire terms; whether existing employees receive equivalent clarity remains, for now, the open question raised by the people who work there.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
