Anthropic insider who quit over apocalypse fears doubles down on doomsday stance: ‘Frighteningly real’

Jacob Coxon, a former Anthropic researcher who resigned this week after three years across Anthropic and OpenAI, told NBC News' Tom Llamas on Wednesday that "there is a substantial probability that this technology could kill everyone," calling the risk "frighteningly real" and not a marketing stunt. Coxon said today's AI models pose "no risk of extinction" but warned on CNN that recursive self-improvement — AI systems conducting their own research — could arrive as soon as 2027 or 2028, with some in the field believing the transition could take roughly six months, and that an unchecked developer race could produce an AI takeover by the end of the decade. He called for international cooperation and regulation, saying the safety problem is "eminently solvable" if governments and AI companies slow the competitive rush.

Anthropic insider who quit over apocalypse fears doubles down on doomsday stance: ‘Frighteningly real’ See more of our coverage in your search results. Add The New York Post on Google https://www.google.com/preferences/source?q=nypost.com Former Anthropic insider Jacob Coxon doubling down on his warning that artificial intelligence could become an extinction-level threat to humanity https://nypost.com/2026/09/09/tech/experts-warn-ai-could-kill-us-all-this-is-how-an-apocalyptic-doomsday-scenario-could-play-out/ within just a few years — insisting his doomsday prediction is no publicity stunt. “There is a substantial probability that this technology could kill everyone,” Coxon told NBC News’ Tom Llamas on Wednesday after resigning from the AI giant this week. “This isn’t hyperbole,” he said. “It’s not a marketing stunt.” Coxon, who spent three years doing research across Anthropic and OpenAI, has sounded the alarm in a blitz of television interviews following his resignation, arguing that rapidly advancing AI could soon become capable of improving itself https://nypost.com/2026/09/09/tech/anthropic-researcher-abruptly-quits-warning-some-developers-believe-ai-could-kill-us-all/ with diminishing human involvement. While Coxon stressed that today’s AI models pose “no risk of extinction,” he told CNN’s Anderson Cooper https://transcripts.cnn.com/show/acd/date/2026-09-09/segment/01 that the danger could change dramatically as soon as 2027 or 2028. The pivotal moment, according to Coxon, would come when AI systems become capable of performing AI research themselves — allowing one generation of models to help create increasingly powerful successors. That feedback loop could trigger what Coxon called an “intelligence explosion,” potentially producing systems vastly smarter than humans. Asked by Cooper how such technology could actually wipe out humanity, Coxon said the scenario sounds like science fiction but is “frighteningly real.” He pointed to possible attacks involving “hacking critical infrastructure” and “extinction-level bioweapons” as examples of how vastly more capable autonomous systems could translate their intelligence into real-world destruction. Coxon said current AI systems may already be capable of causing serious cyber or infrastructure damage, but lack the capability required to outsmart humanity on an extinction-threatening scale. His concern is how quickly that could change. In a separate interview with Fox News’ Bret Baier, Coxon said the immediate threat of an AI takeover was “minimal,” but warned that recursive self-improvement could arrive as soon as next year. Some people working in the field believe the transition could happen in roughly six months, he told Baier. Coxon warned that an unchecked race among developers could ultimately produce an AI takeover by the end of the decade. He called advanced AI “possibly the most dangerous technology that humanity has ever created.” Despite the dire warnings, Coxon said catastrophe is not inevitable. Instead, he argued that AI companies are trapped in a race in which even developers genuinely worried about the technology fear slowing down because competitors — including rival companies or countries — could forge ahead. On CNN, Coxon said AI developers were “begging to be regulated.” He made a similar plea on Fox, calling for international cooperation so countries and companies “don’t race ourselves into extinction.” Coxon said the safety problem itself was “eminently solvable” if governments and AI companies can find a way to slow the competitive rush toward ever-more-powerful systems.