OpenAI safety staffer gives 70% extinction odds in 3 years

Marcus Williams, a member of OpenAI’s Safety Oversight team, wrote on X on September 10, 2026 that he estimates a 70% chance of human extinction within three years unless AI labs slow development or regulators intervene. Williams joined OpenAI in May 2025.

Williams posted a warning and supplied the numerical estimate after another user turned the claim into a wager. The other user offered $10,000 in Microsoft shares against OpenAI stock over five years; Williams responded with a percentage and posted: “70% in the next 3 years if there isn’t regulation/slowdown although i think regulation/slowdown is very possible.” He framed the figure as conditional on whether regulation or a coordinated slowdown occurs.

The post came after a series of developments that drew political attention to advanced AI risks. On September 9 a researcher resigned from Anthropic and accused Anthropic and OpenAI of taking dangerous risks with powerful systems. U.S. lawmakers reacted within a day and a Senate proposal was introduced that would ban superintelligence as part of a broader push for a pause and new legal controls. The U.N. human rights chief described advanced AI as posing an existential risk during a speech in Geneva earlier in September.

Williams’s 70% estimate is higher than several other public assessments from AI researchers. Anthropic’s alignment lead, Evan Hubinger, has placed the odds above 10% for this decade. Nobel laureate Geoffrey Hinton has offered a 10% to 20% range. Market and prediction platforms began incorporating the new estimate into trading and event prices shortly after the post.

Williams wrote that regulation or a coordinated slowdown between labs remains possible and would reduce the risk. The post reflects divergent assessments within the field and added a high-probability estimate to a broader public debate about near-term AI governance.