Former Deepmind PR staffer says the lab once banned public discussion of AI extinction risk

Details

Vishal Maini, who worked on Google DeepMind's communications and policy team from 2018 to 2022, says the lab maintained a blanket prohibition on external discussion of AI-driven human extinction risk. 'External communication about the possibility of human extinction was not permitted, by anyone, at any level of the organization,' he states. Researchers were coached to compare concerns to science fiction — specifically Terminator films — and redirect to positive applications in healthcare and climate. The internal reality was different. Maini says the team knew the AI alignment problem was not solved and that 'there were far too few people working on the problem.' After months of internal pushback, the ban was partially relaxed: positively framed safety content was permitted, including a blog post that discussed extinction risk in accessible, non-alarming language. The account tracks with the direction of travel: growing numbers of AI safety researchers, including those inside DeepMind, are now publicly raising alarms about unsecured models and systems that could slip beyond human control. Maini attributes the narrowing gap between internal knowledge and public messaging to evidence that is 'harder to dismiss now.' The disclosure matters because it documents a structural pattern at one of the world's leading AI labs during a period when safety research was accelerating internally but suppressed publicly — directly relevant to ongoing debates about transparency obligations for frontier AI developers.