Google DeepMind’s Lifetime Gag Clauses: What Changed, What Didn’t

Last updated: 5 November 2026. Figures and quotes below are sourced to primary publications linked inline.

The 60-second version

• The Washington Post reported on 3 November that some Google DeepMind employees in London are bound by non-disparagement clauses that purport to stop them criticising the company publicly for the rest of their lives, even after they quit.
• Google told the Post it has removed the post-termination part of the clause from new-hire contracts and will not enforce it against people who have already left — but it has not said it is amending the contracts current employees already signed.
• The revelation followed a wave of DeepMind resignations over the company’s military AI contracts, including one engineer who told the BBC she defied her own contract to speak out.
• Two of the people who signed a 2024 “Right to Warn” letter demanding frontier labs scrap exactly this kind of clause are reported to work, or to have worked, at DeepMind — which is why critics say this is a promise the industry already made once before.

What the Washington Post found

According to reporting by Gerrit De Vynck for The Washington Post, published 3 November, some Google DeepMind staff in London — a major hub for the lab — have employment contracts containing non-disparagement clauses that are, on their face, intended to apply for life, including after the employee leaves. The clauses bar staff from criticising Google publicly, which current employees told the Post discourages them from discussing the risks and impacts of the AI systems they help build.

No outlet has yet published the contract text itself. What is on record is the reaction: dozens of employees reportedly joined an internal discussion thread about the clauses, some reacting with a “1984” emoji. One current employee, granted anonymity to avoid retaliation, put the problem in practical rather than legal terms: “If it’s not enforceable then the reason is to basically intimidate people.” Another said plainly: “The public wants to know and deserves to know.”

How employees found out: a resignation over military AI, not a routine audit

The contracts didn’t surface from a compliance review — they surfaced because people started checking what they’d actually signed. The trigger was a separate but related controversy: in reporting by Sam Biddle for The Intercept and a BBC Eye investigation by Christopher Giles, both published 21 October, DeepMind research engineer Sofia Liguori resigned over Google’s military contracts with the US and Israel, including Project Nimbus and a classified Pentagon AI contract. In her resignation letter she wrote that she “no longer believe[s] that our technology will have a net positive impact on society,” and told reporters that “the company was very effective at creating a pervasive sense of fear, which induces self-censorship.” A Google spokesperson, Tom Hewett, said the company “follow[s] a consistent process for reviewing and entering these contracts.”

A second researcher, Andreas Kirsch, resigned around the same time, saying Google “has no right to control or veto how the Pentagon uses its AI” once it’s deployed, and arguing that current AI models “are not robust enough” for military use. It was after Liguori said publicly that she had ignored her own contract’s restrictions to speak out that other London-based staff began reviewing their own agreements and discovered the lifetime clause. The gag-clause story and the military-AI story aren’t two unrelated items that happened to land in the same week; the second produced the first.

What Google says changed — and what evidently hasn’t

Google’s statement to the Post says: “We’ve always encouraged an open dialogue about safety and responsibility,” and that it “updated our terms for new hires, which included removing the post-termination aspect of non-disparagement clauses.” Separately, the Post reports a Google executive told staff internally, the Friday before the story ran, that the company would not enforce the clause against people who have already left.

Read those two statements together and a gap opens up. “Removing the post-termination aspect… for new hires” changes the template going forward; it says nothing about contracts current employees are still bound by. “Will not enforce it” is a promise about Google’s conduct, not a change to the document — legally weaker, and far easier to reverse, than striking the clause outright. Several aggregator write-ups compressed this into headlines like “Google ends non-disparagement requirements,” which overstates it: as of this writing, existing employees’ contracts are reported unchanged, with no amendment published or offered for signature.

That distinction matters because it’s exactly where UK whistleblower law sits. Britain’s public-interest disclosure protections could, in principle, override a gag clause for a protected disclosure — but protections a tribunal vindicates after the fact do little to stop an employee self-censoring in the moment, which is the harm employees describe.

The déjà vu: a promise the industry already made in 2024

This is not the first time a frontier lab’s non-disparagement practices have become the story instead of its models. In May 2024, OpenAI was revealed to be threatening departing employees with the loss of vested equity if they declined to sign broad non-disparagement and confidentiality terms; Sam Altman apologised publicly, and OpenAI said it would release former employees from those obligations and strike the clawback provision.

That controversy produced the “Right to Warn” open letter, published 4 June 2024 by current and former employees of OpenAI and Google DeepMind and endorsed by Yoshua Bengio, Geoffrey Hinton and Stuart Russell. Its first demand was explicit: that AI companies commit “not [to] enter into or enforce any agreement that prohibits ‘disparagement’ or criticism of the company for risk-related concerns, nor retaliate for risk-related criticism by hindering any vested economic benefit.” Two signatories are named as current or former Google DeepMind staff — Ramana Kumar and Neel Nanda. That a lifetime version of exactly the clause the letter sought to abolish was still sitting in London contracts two and a half years later is the detail that gives this story its edge: it suggests the 2024 commitments functioned as a public-relations reset rather than a change to the paperwork.

The pattern sits inside a wider year of voluntary-commitment scrutiny. CurratedBrief has covered how a New York City Council hearing exposed gaps between labs’ voluntary safety pledges and their actual practice, how the FTC opened a safety probe into OpenAI and Anthropic because self-regulation was being treated as sufficient, and how Brussels opened its first AI Act enforcement cases against the same two labs. Non-disparagement clauses are a narrower mechanism than any of those regulatory fights, but they work on the same lever: whether anyone outside the company hears about a problem before a regulator, journalist, or court forces the issue.

Why this isn’t just a Google story

Every frontier lab relies on self-reported transparency: safety evaluations, voluntary disclosures, red-teaming write-ups. That rests on an assumption that researchers are free to say when something looks wrong. A clause that chills that speech — whether or not it would survive a courtroom challenge — degrades the reliability of everything a lab says about itself, including the true parts. It’s also a competitive signal: Google DeepMind, whose co-founder Demis Hassabis joined the chief executives of OpenAI and Anthropic in a September call for a measured approach to frontier AI development, is now the subject of exactly the kind of internal-culture story that undercuts a “trust us to be careful” pitch — the same tension CurratedBrief noted when Google limited early access to its Gemini 4 Carbon benchmark to cyber-defence specialists rather than release it broadly: the company asks to be judged on its caution, while a contractual document asks employees to stay quiet regardless of what they actually see.

What this means if you build, publish, or run software

If you negotiate technical hiring contracts, check whether your own template carries a post-termination non-disparagement or confidentiality clause with no time limit and no carve-out for safety-related disclosures — and whether that carve-out is actually in the document, not just a verbal assurance to candidates, who increasingly ask.

If you’re the one signing, read those sections before you need to rely on them, and treat an employer’s informal “we won’t enforce that” as exactly what it is: present intent, not a release. If it matters to you, ask for the clause struck or time-limited in writing before you sign, not promised afterward.

If you cover this industry or lean on labs’ self-reported safety claims in your own product decisions, separate two categories coverage often conflates: a company’s public statement of intent, and its actual contract or policy document. They diverge more than headlines suggest, and the gap is usually where the real story lives — as with California’s new AI auditor registry under SB 813, which exists precisely because regulators stopped taking lab self-reporting at face value.

What we still don’t know

• The exact wording of the clause. No outlet has published the contract text, so claims about its precise scope rest on paraphrase.
• How many employees are affected, in London or elsewhere. Reporting says “some” staff and “dozens” in an internal thread — not a verified total, and not confirmation the clause is UK-only.
• Whether Google will issue a written amendment to existing contracts, or whether “we will not enforce it” is the final position. The two are legally different, and only one survives a change of management.
• Whether this practice exists at other frontier labs’ UK or EU entities under similar wording — nothing in the current reporting establishes that either way.
• Whether UK whistleblower protections would hold up against this specific clause if tested in a tribunal; no such case has been brought.

FAQ

Did Google DeepMind force employees to sign lifetime gag orders?
Some current and former London employees say their contracts contain non-disparagement clauses that read as applying for life, including after departure. Google hasn’t disputed the clauses exist; it disputes how they’ll be applied.

Has Google actually removed the clauses?
Only from new-hire contracts, by its own account. For existing staff, Google says it won’t enforce the post-termination restriction — a policy position, not a documented contract change.

Is a lifetime non-disparagement clause enforceable in the UK?
Untested. UK public-interest disclosure law can protect specific disclosures, but that legal backstop doesn’t neutralise the chilling effect employees describe, and no tribunal has ruled on this clause.

How does this compare with OpenAI’s 2024 controversy?
Similar shape — broad terms discovered by departing staff, public clarification, a fix presented as policy — but OpenAI’s case involved an explicit threat to claw back vested equity, which doesn’t appear in the DeepMind reporting so far.

Sources

• Gerrit De Vynck, “Google researchers push to end non-disparagement clauses,” The Washington Post, 3 November 2026 (syndicated)
• Christopher Giles, BBC Eye investigation, 21 October 2026
• Sam Biddle, “Google DeepMind Engineer Resigns in Protest of Pentagon Contract,” The Intercept, 21 October 2026
• “A Right to Warn about Advanced Artificial Intelligence,” 4 June 2024
• Lee Wan-ki, “Google DeepMind Workers Push Back on Lifetime Gag Clause,” Seoul Economic Daily, 4 November 2026

Please follow and like us: