AI researchers put human extinction risk as high as a coin flip in new video interviews

Researchers speak out in collected interviews

Palisade Research, a nonprofit that describes its work as studying the capabilities and motivations of AI systems, has published a series of video interviews with AI researchers on a site called frominside.ai. According to The Verge, the collection includes about a dozen interviews with current and former employees of OpenAI, Google and Anthropic, and several of them warn that advanced AI could plausibly drive humans to extinction.

The bluntest assessment comes from Geoffrey Irving, who has worked at both OpenAI and Google DeepMind. In his interview, he put the chance of human extinction from AI at "about a coin flip."

Risk estimates vary but stay high

Neel Nanda, a research scientist at Google DeepMind, gave a lower figure while making a similar point, saying there is "at least a 10 percent chance" that AI causes human extinction, a probability he called "ridiculously high."

Daniel Kokotajlo, a former OpenAI researcher, framed the problem in terms of control. A superintelligent system, he argued, would be essentially "god-like powerful," and because researchers do not know how to control such systems, nobody would end up controlling them. The situation, he said, is "exactly as dangerous as it sounds and must not be allowed to happen."

Motives and skepticism addressed head-on

The site does not present the interviews in isolation. According to The Verge, it also collects segments of multiple researchers answering shared questions, including whether the warnings amount to hype or marketing, and why anyone would keep working on technology they consider this dangerous.

The answers vary in candor. Nanda said he would quit if he did not believe his work was directly reducing existential risks, adding that the companies building these systems are not going to stop just because one researcher resigns. Mary Phuong of Google invited suspicion of her own incentives, saying people "absolutely should be suspicious" of her statements because the lab pays her.

No shared plan

The Verge notes that it had previously reported on comments from Nanda and Irving in a broader piece about the "AI safety" community, and about how difficult it is to pin down what that term means or what anyone should do about it. The new collection does not settle the question: according to The Verge, the videos do not offer a single cohesive solution.

Why it matters

These warnings come not from outside critics but from people who build frontier AI systems, some of them still employed at the major labs. When researchers with insider access publicly put extinction odds anywhere between 10 percent and a coin flip, it signals that catastrophic risk is a live concern within the field rather than a fringe position. At the same time, the absence of any unified remedy, combined with an open acknowledgment that lab employees have financial incentives, leaves companies, regulators and the public without a clear way to judge or act on the warnings. How the labs respond to their own researchers' estimates may shape safety policy, investment decisions and public trust as these systems continue to roll out.