Former Anthropic researcher warns AI labs aim to automate research, urges slowdown

Jacob Coxon, a 27-year-old former capabilities researcher at Anthropic who resigned last month, testified before the New York City Council that automating AI research has become the main objective of leading frontier AI companies. He said that at OpenAI milestones were set for an automated AI researcher by 2027-2028 and that the timeline is now “months from now, not decades.”

Coxon warned that once AI systems can conduct their own research, “humans will be much further out of the loop than we are today, and today is already not great.” He noted that most code today is generated by AI and that developers often do not review it carefully.

The hearing, which also featured representatives from OpenAI, Anthropic, Google and Meta, was described by City Council speaker Julie Menin as intended to ensure responsible innovation rather than to hinder progress. Coxon urged the council to demand more transparency from AI labs and to consider slowing the pace of development, arguing that a “move fast, break things” approach is unsuitable for powerful technology.

Anthropic CEO Dario Amodei later published an essay of nearly 4,000 words calling on the industry to “slow the pace at which we improve the capabilities of AI models,” emphasizing the need to use any gained time wisely. OpenAI and Anthropic did not immediately respond to requests for comment.