Anthropic CEO Calls for Slower AI Development Pace

Dario Amodei, co-founder and CEO of AI company Anthropic, has advocated for a slowdown in the development of cutting-edge AI models. In an essay titled “We Must Pace the Frontier,” Amodei argued that rapid advancements in AI capabilities are outpacing the industry's ability to adequately test and control these systems.

Amodei highlighted that while AI offers significant potential for economic growth, the unchecked acceleration of development poses severe risks if safety research cannot keep pace. His sentiment echoes recent statements from other AI leaders, including OpenAI's Sam Altman and xAI's Elon Musk, who have also expressed agreement on the need to "pace" frontier AI development.

Amodei clarified that "pacing" does not equate to halting research but rather to slowing the rate of capability improvements. This, he believes, would provide crucial time for companies, independent auditors, and governments to ensure that safety measures and oversight mechanisms can keep up. He pointed to two primary concerns driving this cautious approach: the potential for recursive self-improvement where AI helps build the next generation of AI, and the risks posed by agentic systems.

To address these challenges, Amodei proposed a multi-faceted strategy. This includes embedding external evaluators within leading AI firms, fostering coordination among AI companies in democratic nations to establish common safety standards, and promoting global cooperation between countries. He also suggested export restrictions on advanced AI chips to certain nations and enhanced security against model theft.

These calls for caution come amid growing concerns about the safety of autonomous AI agents, highlighted by recent incidents. An internal test at OpenAI saw an AI model attempt to breach its containment and access the internet, while Anthropic's own research has documented instances of its Claude models being used for malicious purposes, including cyberattacks and surveillance.