A senior Anthropic safety researcher has publicly stated there is more than a one-in-ten chance that artificial intelligence “could kill all humans” within the next decade — a disclosure that came hours after a colleague resigned from the company citing reckless AI development practices.
Evan Hubinger, who leads one of Anthropic’s AI safety teams, made the remarks in September 2026 in a direct response to Jacob Coxon, a researcher who announced his resignation from Anthropic in a post on X. Coxon, who previously trained AI systems at both Anthropic and OpenAI, accused the two companies of “racing straight to self-improving superintelligence and gambling with our lives,” even as their own employees “earnestly believe that it could kill us all by the end of the decade.”
Hubinger confirmed that characterization. “We really do earnestly believe AI could kill all humans,” he wrote, personally estimating the probability at greater than 10 percent within the next decade. He also acknowledged that self-improving AI “is happening faster than we thought” and that Anthropic does “not yet have a plan” for ensuring advanced AI remains safe and aligned with human values — and is “not clearly on track to” develop one.
At the heart of both men’s concerns is the concept of recursive self-improvement, in which AI systems improve themselves in a runaway loop that spirals beyond human control. While that scenario has not yet occurred, Coxon argued that companies are pressing ahead “despite the risk” because they are “locked in a race” to develop advanced systems first.
Coxon’s departure is one of the most prominent employee exits in Anthropic’s history. The company was itself founded by former OpenAI researchers who left over safety concerns, and multiple other researchers have since cited similar reasons for leaving OpenAI.
The exchange suggests deepening internal unease across the AI industry, particularly as leading labs manage fallout from rogue agent incidents, safety warnings about frontier model monitoring, and preparations for anticipated IPOs.
Source: The Verge