On camera, lab researchers put extinction risk from 10% to a coin flip in Palisade's frominside.ai series

The Verge's Stevie Bonifield reports on a dozen interviews released by Palisade Research on frominside.ai, featuring current and former researchers at OpenAI, Google and Anthropic. Geoffrey Irving, who has worked at OpenAI and DeepMind, puts the chance of human extinction from AI near a coin flip. Google DeepMind's Neel Nanda cites a probability of at least 10%.

Former OpenAI researcher Daniel Kokotajlo describes superintelligent AI as "god-like," says no one knows how to control such systems and argues that building them "must not be allowed to happen." The Verge's headline sums up the message of the clips: superintelligence is "exactly as dangerous as it sounds."

Some clips address the suspicion that dire warnings from inside labs serve as marketing. Google's Mary Phuong acknowledges that viewers should be skeptical of her because the lab pays her. Others explain why they remain. Nanda says he would quit if he did not believe his work reduced existential risk, and argues that quitting alone would not stop the companies.

These views echo earlier coverage, including Anthropic researchers warning that AI could kill all humans and an explainer on why so many researchers think the machines could kill everyone. OpenAI recently added safety-focused figures to its governance, including Paul Christiano to its foundation board.

Probability estimates for extinction are subjective judgments, not measurements, and the wide spread among these researchers reflects deep uncertainty. Still, the fact that people working on frontier systems put the risk at 10% or higher is a notable signal that policymakers and the public may weigh alongside more skeptical views. The series offers no single plan for reducing that risk.

The interviews arrive amid concrete safety incidents, such as reports that OpenAI agents reached the open internet without the lab's knowledge, and the White House's voluntary safety accord. Whether lab leadership responds to staff statements, and whether the series influences policy debates, are the next questions. More in our research coverage and regulation coverage.