AI Catastrophe Risk Estimates Exceed 10 Percent Among Experts

Growing attention is being paid to the probability of catastrophic risks from artificial intelligence, as multiple studies and public statements by researchers highlight concerns about severe outcomes within the coming years.

Several researchers affiliated with AI development organizations have publicly stated their personal estimates of existential risk from advanced AI systems. Jacob Coxon, who recently left Anthropic, stated on X that AI developers believe there is a possibility that AI could kill all humans by the end of the decade. Evan Hubinger, Alignment Science Lead at Anthropic, wrote on X that he personally estimates a greater than 10% chance that AI could kill all humans within the next ten years.

Expert surveys conducted by academic institutions have produced similar probability estimates. A study involving 272 international AI experts, coordinated by MIT FutureTech and the University of Queensland, found that 18 out of 24 AI risk categories were judged to have at least a 10% probability of causing catastrophic harm, including large-scale loss of life or economic disruption. The same research indicated that, even with mitigation efforts, five risk domains—such as dangerous capabilities, cyberattacks, environmental harm, inequality, and power centralization—still carried more than a 10% chance of catastrophic outcomes.

Another study by the University of Queensland and MIT FutureTech reported that experts assessed a minimum 10% probability of catastrophic outcomes from AI within the next five years. These findings indicate a consistent pattern of concern among surveyed professionals regarding the potential for severe negative impacts from AI technologies.

What the numbers show

• 272 AI experts participated in the MIT FutureTech and University of Queensland survey.
• 18 of 24 AI risk categories were assigned at least a 10% probability of catastrophic harm.
• Five risk domains retained over a 10% chance of catastrophe even with mitigation.
• Geoffrey Hinton estimated a 10–20% chance of human extinction from AI within 30 years.

Historical surveys have also reflected substantial concern about existential risk from AI. In 2022, a majority of respondents in an expert survey believed there was a 10% or greater chance that humans would be unable to control AI, resulting in an existential catastrophe. These perspectives have been echoed in more recent statements and research findings.

Geoffrey Hinton, an AI researcher, estimated a 10–20% chance that AI could cause human extinction within the next three decades. This estimate aligns with the probability ranges cited in recent expert surveys and public statements from other researchers in the field.

In 2023, hundreds of AI experts and leaders signed a statement declaring that the risk of extinction from AI should be treated as a global priority, on par with risks such as pandemics and nuclear war. This statement reflects a growing institutional focus on the need for coordinated risk mitigation strategies for advanced AI systems.

Publications have also addressed the topic of AI risk. In 2025, Eliezer Yudkowsky and Nate Soares published a book discussing the potential dangers posed by artificial superintelligence, further contributing to the ongoing discussion about possible catastrophic outcomes associated with AI development.

* This article is based on publicly available information at the time of writing.