AI researchers warn superintelligence extinction risk is around 50 percent

By Splitfeed EditorialLast updated Sep 29, 2026, 17:41 UTC

On September 29, 2026, Palisade Research, a non-profit that studies AI capabilities and motivations, published a series of roughly a dozen video interviews on its site frominside.ai. The interviews feature current and former employees from OpenAI, Google, and Anthropic speaking about the dangers of superintelligence. The most specific estimate came from Geoffrey Irving, who has held roles at both OpenAI and Google DeepMind, and who stated that he puts the chance of human extinction from superintelligence at approximately 50 percent, describing it as roughly a coin flip. Palisade Research framed the project with the explicit claim that superintelligence is exactly as dangerous as it sounds.

The publication matters because coin-flip extinction estimates from people with direct, senior experience inside the leading AI labs are rare on the record. Most public warnings from insiders arrive as vague cautions or carefully hedged policy language. Irving's framing is unusually concrete, and the fact that it appears alongside similar warnings from researchers still affiliated with OpenAI, Google, and Anthropic gives the series a collective weight that a single defector's statement would not carry. What is genuinely in dispute is whether these researchers represent a meaningful consensus inside the industry or a vocal minority, and whether public warnings of this kind accelerate safety work or simply alarm without providing actionable direction.

No reactions from the Pro-AI, Anti-AI, or Middle Ground camps had been published at the time of writing. The Pro-AI camp would typically argue that extinction-level probability estimates are speculative, that current systems are far from the kind of superintelligence being described, and that alarmism distracts from productive development. The Anti-AI camp would likely treat Irving's 50 percent figure as confirmation that the industry itself recognises a civilisational risk it is failing to adequately restrain. The Middle Ground camp would generally call for taking the warnings seriously as a policy input while pushing back on the precision of any single probability estimate as a guide to regulatory action.

The claims in the series will be tested by how major labs respond publicly, whether any regulatory bodies cite the interviews in upcoming legislative proceedings, and whether additional current employees follow with similarly specific on-the-record statements in the months ahead.