An OpenAI safety researcher put extinction odds at 70%, then the post vanished

A safety researcher inside OpenAI put a number on the worst case, 70 percent within three years, and then the posts disappeared from X.

What was posted

On September 10, 2026, Marcus Williams, a member of technical staff on OpenAI’s safety oversight team, posted that unless there is AI regulation or a coordinated slowdown between labs, “human extinction in the next few years seems very likely.” When another user pushed him for a number, he replied, as reported by BeInCrypto: “70% in the next 3 years if there isn’t regulation/slowdown although i think regulation/slowdown is very possible.”

Both posts have since been deleted from his account, while the quotes live on in the coverage. The hedge matters as much as the number: Williams framed the figure as conditional and called regulation or a slowdown “very possible.”

How far outside the range this sits

Williams posted in a personal capacity, and OpenAI has not commented. The figure is far above most public AI risk forecasts. Geoffrey Hinton has cited a range of 10 to 20 percent. Evan Hubinger, who leads alignment science at Anthropic, puts the odds above 10 percent this decade.

The post traveled fast: Polymarket pushed it to its followers within hours, per BeInCrypto. Days later, the heads of Anthropic, OpenAI and xAI publicly backed slowing down frontier AI development, and the estimate resurfaced as part of that debate.

Why an uncheckable number still matters

The figure itself cannot be verified, and a deleted post cannot be interrogated. What can be observed is who produced it: someone whose job is monitoring safety inside a frontier lab. When estimates like this come from inside the companies building the systems, the AI safety debate stops being a research argument and becomes a political one.