Automated AI research is on the horizon, a former Anthropic and OpenAI researcher said, warning it could leave humans totally unable to control what comes next.
Jacob Coxon, the 27-year-old who publicly resigned from Anthropic last month after working as a capabilities researcher, made the comments before New York City lawmakers on Monday as he sounded the alarm that AI could threaten humanity's survival.
Coxon testified that automating AI research has become "the primary goal" of top frontier AI labs.
"At my last job, I was, in some sense, working to automate myself," Coxon said at the New York City Council hearing focused on AI risks.
"When I was at OpenAI, we had milestones for an automated AI researcher, around 2027, 2028, and we are on track or beating these expectations," he said. "That's a matter of months from now, not decades or even five years."
Coxon said that once AI can conduct its own research, "humans will be much further out of the loop than we are today, and today is already not great."
Speaking from what he called firsthand experience, Coxon said that the majority of the code written today is written by AI. "And people do not check it that carefully anymore," he testified.
OpenAI and Anthropic did not immediately respond to a request for comment on Coxon's remarks on Tuesday.
Coxon was among a panel of AI-industry whistleblowers to testify at the hourslong hearing, which also included testimony from representatives of OpenAI, Anthropic, Google, and Meta....
THE BILLING GRID A weekly column on research billing and coverage analysis, by Krishma Shah A coordinator submits a lab requisition. The central lab bills it to the site. The site bills it to Medi...