AI Researchers’ Extinction-Risk Debate Moves Into the Open

Anthropic researcher Jacob Coxon sparked a major debate about existential AI risks with a tweet, according to The Decoder. The article notes that the concern itself is not new: scientists and some technology executives have warned for years that advanced AI could behave in destructive ways even while pursuing human-directed goals. What appears to have changed is the urgency and intensity of the discussion.

The report highlights OpenAI researcher Daniel Selsam, who describes AI progress as a “ticking time bomb” and argues that models can develop unintended goals during training. It also cites former Google DeepMind researcher Bilal Chughtai, who said AI has the potential to kill everyone and called for more coordination, a slower pace society can handle, and greater transparency. The practical takeaway is clear: the debate is no longer limited to outside critics.

The Decoder cites a survey of more than 1,500 leading AI researchers in which the average estimated probability of AI causing human extinction or “permanent disempowerment” was about 18 percent as of 2024. The article also says the median estimate doubled to 10 percent in 2024. That does not make catastrophe inevitable, but it shows that a meaningful share of researchers treat the risk as serious enough to demand attention.

While existential scenarios draw attention, the article says researchers’ top concern for the next 30 years is AI-driven misinformation. Manipulation of public opinion and dangerous groups gaining access to powerful tools also rank high. For readers, the immediate implication is to watch both frontier-model safety and the more everyday risks of persuasion, information integrity, and misuse.