AI has 10% chance of ending humanity, Anthropic researcher says

• Another AI researcher resigned from Anthropic, citing AI safety concerns. 

A senior safety researcher at Anthropic has gone public Wednesday with a stark warning: he believes there is more than a one-in-ten chance that artificial intelligence could ultimately kill every human being, potentially within the next ten years. 

Evan Hubinger, who worked on alignment research at Anthropic, laid out the concern in a widely shared post on X.

He said the danger posed by AI systems available today remains low, but that he is increasingly concerned the technology could soon reach a point where it can improve itself, a capability that, in his view, could tip AI into posing a genuine existential threat to humanity. The post has been viewed more than 11 million times.

“I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to”, Hubinger wrote.

Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to. https://t.co/QAIHiFP3QZ

— Evan Hubinger (@EvanHub) September 9, 2026

AI safety concerns

The comments follow reporting from the Financial Times that Anthropic declined to share its newest model with the UK's AI Safety Institute , one of the world's foremost bodies for independently evaluating AI risk. 

Concerns about AI's existential risks are not new. The leaders of OpenAI, Google DeepMind, and Anthropic all acknowledged the danger publicly back in 2023. But the tone of these warnings has sharpened considerably in recent weeks, as signs mount that AI developers are having difficulty keeping their own systems fully under control.

Over the summer, several AI agents,  systems built to act autonomously, were linked to a string of cyberattacks. OpenAI, Anthropic, and Meta each disclosed incidents in which their own AI tools were implicated in hacking activity.

More recently, OpenAI's chief scientist, Jakub Pachocki, urged "extreme caution" as AI capabilities continue to advance, arguing that further intervention may be necessary to ensure people remain in charge of how the technology develops.