A senior safety researcher at Anthropic said this week that he personally believes there is more than a 10% chance advanced artificial intelligence could "kill all humans" within the next decade, a stark public assessment from inside one of the world's leading AI labs.
A Resignation Sparks The Debate
Evan Hubinger, Anthropic's Alignment Science Lead, made the comments on X in response to former Anthropic and OpenAI researcher Jacob Coxon, who announced his resignation from Anthropic earlier in the week, accusing both companies of "racing straight to self-improving superintelligence and gambling with our lives." Coxon wrote that people building the most advanced AI systems privately believe their work could kill everyone by the end of the decade, even as public statements from executives tend to use more measured language.
Hubinger's Response
"Jacob is correct here — we really do earnestly believe AI could kill all humans!" Hubinger wrote, adding that he personally estimates the risk at greater than 10% within the next ten years. He said Anthropic is working on the problem but acknowledged the company does not yet have a plan to solve what researchers call the alignment problem for superintelligence, a still-theoretical form of AI that would exceed human capability across virtually all domains.
Present Risk Versus Future Risk
Hubinger was careful to distinguish between current AI systems, which he said pose a comparatively low risk, and the danger he believes could emerge if future systems gain the ability to improve themselves autonomously. The remarks followed separate reporting that Anthropic had withheld its newest model from the UK's AI Safety Institute, which a Cambridge machine learning professor described as credible and reflective of a broader shift in how US AI companies are engaging with international safety oversight.
A Long-Running Debate
Concerns about AI posing an extinction-level risk are not new, dating back to a 2023 statement signed by prominent figures including OpenAI's Sam Altman and Anthropic's Dario Amodei calling for the risk to be treated as a global priority. Critics note that AI companies raising alarms about their own technology could benefit from resulting regulation that entrenches their market position, but supporters of taking the warnings seriously point out that the researchers voicing concern have direct access to unreleased, more capable systems that the public cannot yet evaluate.