Ex-DeepMind spokesperson says lab barred staff from publicly discussing AI extinction risk

A former Google DeepMind communications staffer says the lab once barred its employees from publicly discussing the prospect of human extinction driven by artificial intelligence. Vishal Maini, who worked in public relations at the London-based lab, said the restriction applied to everyone in the organization regardless of seniority. The claim, reported by The Decoder, offers a rare look at how one of the world's leading AI companies managed its public posture on its own most alarming risk scenario.

Maini described the policy in blunt terms: external communication about the possibility of human extinction was not permitted, by anyone, at any level of the organization. The rule was not a matter of individual caution but a blanket prohibition, according to his account. That distinction matters because DeepMind has long positioned itself as a safety-focused lab, publishing research on alignment and signing prominent statements on existential risk.

Internally, Maini said, staff understood that the technical problem of aligning advanced AI systems with human values had not been solved. That gap between private understanding and public messaging is the core of his allegation. If accurate, it suggests DeepMind's cautious public language during that period reflected a deliberate communications strategy rather than a settled scientific consensus inside the building.

The timing of the alleged policy is notable. DeepMind was acquired by Google in 2014 and spent years building its reputation partly on safety research, including work on specification gaming and reward hacking. Several of its own researchers have since signed open letters warning that mitigating extinction risk from AI should be a global priority alongside pandemics and nuclear war. A lab that cultivates that image while restricting staff from voicing the same concern publicly faces an obvious credibility question.

The claim also lands in a crowded field of former-employee accounts. Ex-staffers at OpenAI, Anthropic and other labs have gone public in recent years with concerns about internal safety culture, sometimes after signing nondisclosure agreements. DeepMind has not publicly responded to Maini's characterization, and the company's current communication policies may differ from those he describes.

What remains unclear is how formal the restriction was and when it lapsed. Maini's account is a single former employee's recollection, and no internal document has surfaced to confirm a written policy. The larger question is whether other labs adopted similar rules, and whether the industry's public talk about existential risk matches what its communications teams allow employees to say.