DeepMind Banned AI Extinction Risk Talk, Ex-Staffer Says

Vishal Maini, who worked on DeepMind's communications and policy team from 2018 to 2022, recently shared that the Google-owned AI laboratory previously enforced a strict ban on discussing human extinction risks. According to Maini, external communication regarding the existential threats of artificial intelligence was completely forbidden for all employees, regardless of their rank within the organization.

Instead of addressing these concerns openly, DeepMind researchers were reportedly instructed to downplay existential risks. Maini noted that staff were coached to dismiss such scenarios as mere alarmism, frequently comparing them to Hollywood tropes like the Terminator movies. Instead of engaging with safety critics, employees were directed to pivot public conversations toward the positive societal impacts of AI, such as its applications in healthcare and climate change mitigation.

Internally, however, the sentiment was vastly different. Maini stated that the communications and policy team was well aware that the AI alignment problem remained unsolved. Furthermore, there was a widespread internal consensus that far too few researchers were actively working on solving these safety issues. After months of internal pushback, DeepMind eventually relaxed its policy, allowing positively framed safety discussions, including a blog post that addressed extinction risks using softer, more palatable language.

For AI practitioners and researchers, Maini's revelations underscore the historical tension between corporate public relations and genuine scientific transparency. The disclosure comes at a time when an increasing number of AI safety researchers, including those currently at DeepMind, are publicly warning about the dangers of unsecured frontier models. Maini suggests that the gap between what AI labs know internally and what they share publicly is finally narrowing, largely because the mounting evidence of potential AI dangers has become increasingly difficult for companies to ignore.