IT之家 · 智能时代 · 10/8/2026, 2:15:30 AM
Ex-Anthropic Researcher Warns AI Self-Research Targets Hit 2027-2028, Citing Existential Risk
Former Anthropic and OpenAI researcher Jacob Coxon testified at a NYC council hearing that top labs are targeting fully automated AI research by 2027-2028. Based on internal progress, he estimated the probability of humanity losing control to AI exceeds 50%, criticizing the industry's 'move fast and break things' approach as dangerously inadequate for frontier safety.
SOURCE COVERAGEOriginal coverage
On October 8, Business Insider reported that Jacob Coxon, a former researcher at Anthropic and OpenAI, warned that AI will soon be capable of conducting research autonomously, potentially leaving humans with no control over subsequent developments.
Coxon, 27, previously worked on model capability research at Anthropic and publicly announced his resignation last month. On the 5th local time, he issued a warning to New York City council members, stating that AI could pose an existential threat to humanity.
Testifying at a New York City Council hearing on AI risks, Coxon stated that the primary goal of top frontier AI labs is to automate AI research. "In my previous role, I was essentially researching how to make AI replace myself. At OpenAI, we set milestones for automating AI researchers around 2027 or 2028. Progress has been on track, even exceeding expectations. That timeframe is only months away—not decades, not even five years."
Once AI can conduct research independently, human involvement and intervention in the process will be significantly lower than today’s already suboptimal levels. Based on his firsthand experience, most code is now written by AI, and "people are no longer scrutinizing the code as carefully."
Coxon testified to council members that his tenure at Anthropic and OpenAI convinced him that neither company has adequate safety measures yet continues to push development forward. "The probability that humans lose control and hand it over to AI is greater than 50%, which could ultimately lead to human extinction. Given the magnitude of the consequences, the approach taken by these two companies is extremely reckless."
When Coxon announced his resignation from Anthropic on X, he posted accusations that both Anthropic and OpenAI were acting irresponsibly, gambling with people's lives. The post subsequently went viral.
At the hearing, Coxon urged AI labs to increase transparency and slow down the rapid pace of technological advancement. "These companies are applying startup logic: move fast and fix things later. This might work for a photo-sharing app, but not for developing the most powerful technology in history."
As previously reported by IT Home, when Coxon resigned in September this year, he issued a warning: "The people developing AI genuinely believe that AI could kill us all before the end of the 2020s. This is not marketing hype."