OpenAI Safety Researcher Predicts 70% Human Extinction Risk Within 3 Years

A safety team member at OpenAI (formerly known as OpenAI) has stated that if AI research labs do not slow their pace or if regulators do not intervene, there is a 70% probability of human extinction within three years.
Marcus Williams, who joined the company in May 2025, posted this warning on social media platform X. After initially sharing his message, he later disclosed the figure when another user requested it.
Williams emphasized that without regulation or coordinated slowdowns among research labs, the risk of human extinction within the coming years is very high.
When asked to bet against him on the probability, Williams responded with a ratio rather than a specific number. A different user suggested a wager involving Microsoft stock and OpenAI's five-year performance.
Williams also stressed that while he provided the 70% figure, he did not definitively predict this outcome, noting that regulatory or coordinated slowdown measures could still be possible.
The post came amid heightened political discussion about AI risks in Washington. On September 9, an Anthropic researcher resigned, criticizing both Anthropic and OpenAI for gambling with human survival. U.S. lawmakers quickly responded the next day with calls to pause AI development, including a Senate bill banning superintelligence outright.
Other voices have joined the warning. The UN High Commissioner for Human Rights described advanced AI as an existential risk during a Geneva meeting this week.
However, Williams' assessment is more extreme than others. Anthropic's alignment science lead Evan Hubinger estimates a 10% or higher chance of extinction within a decade, while Nobel laureate Geoffrey Hinton has cited a range between 10-20%.
Crypto traders have also been tracking these warnings closely. Prediction market platform Polymarket quickly disseminated Williams' post to its users, with bettors already evaluating elections and macro events alongside token probabilities.
Ultimately, the near-term test lies more in politics than technology. Whether regulatory bodies will act within the three-year window Williams mentioned remains uncertain, though he indicated this possibility still exists at present.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 오픈AI 연구원, 3년내 인류 멸종 확률 70%