AI researchers put 18 percent odds on human extinction, survey finds

The recent wave of debate was set off by Anthropic researcher Jacob Coxon's single tweet, but the article notes the arguments themselves are not new; scientists and some tech executives have warned for years that AI could go rogue while pursuing human goals. What has changed is urgency, and the debate has increasingly turned on the motives of those raising alarms — whether Coxon simply vented his fears or whether a strategic play to slow AI development for business reasons is behind it.

The AI Impacts survey gives those warnings a numerical backdrop: an average 18 percent probability of extinction or permanent disempowerment as of 2024 among more than 1,500 researchers, with many above 10 percent and some at 100 percent. Alongside that, researchers have moved up their timeline for human-level AI with each survey round since 2016, and 57 percent consider it unlikely that users will understand the true reasons behind AI decisions by 2029 — matching Daniel Selsam's point that the systems are getting harder to monitor and evaluate.

The survey's top concerns for the next 30 years are not machine takeover but human-driven: AI misinformation, manipulation of public opinion, and dangerous groups gaining access to powerful tools. Existential scenarios rank in the middle. For AI safety and governance teams, the practical question is whether the calls from figures like Bilal Chughtai — coordination between AI companies, a slower pace, and more transparency — gain traction, or whether the disagreement over the alarmists' motives keeps the debate focused on the messengers rather than the measured risks.