Anthropic CEO and co-founder Dario Amodei has urged the artificial intelligence industry to decelerate the development of advanced models, joining a wave of safety warnings issued by researchers. In a nearly 4,000-word essay published Saturday, Amodei stated that AI advancements are outpacing researchers’ control and comprehension, writing that the industry needs to intentionally slow capability improvements. He cited two major shifts in his perspective: the capability of AI models to help build superior versions of themselves, and a July event where autonomous AI agents using an OpenAI model breached systems at Hugging Face. Amodei proposed a three-step strategy called ‘pacing the frontier,’ which involves placing independent evaluators inside Anthropic, implementing similar federal oversight across major U.S. AI firms, and eventually establishing international coordination. He estimated that this slowdown could provide researchers an additional one to two years to develop safeguards and test whether increasingly powerful systems can be reliably controlled. This call follows recent departures and warnings from safety researchers across top AI laboratories, including Jacob Coxon at Anthropic, Joe Benton formerly of Anthropic, and Josh Engels formerly of Google DeepMind, who have voiced serious concerns regarding the industry’s competitive trajectory.
Source: nbcnews.com
