AI safety requires more than just slowing our pace | Stuart Russell

Summary

The article examines a proposal by Anthropic CEO Dario Amodei to slow the pace of frontier artificial-intelligence development because of growing concerns about recursive self-improvement and the possible loss of human control over advanced systems. Amodei’s 3,800-word letter, titled “We Must Pace the Frontier,” follows the resignation of Anthropic AI-safety researcher Jacob Coxon and recent controversy surrounding an OpenAI/Hugging Face incident, though the article provides few details about either event. Amodei’s plan has three main elements: independent third-party evaluators embedded within AI companies; common safety standards and limits on unchecked progress among companies in democratic countries; and a wider compact involving authoritarian countries. The proposal does not call for an immediate halt to AI research, but for additional time—possibly one to two years—to improve interpretability, alignment, testing, and safety methods. The author, Stuart Russell, argues that pacing alone is inadequate. He says AI companies should meet defined safety requirements before advancing to systems with more dangerous capabilities, effectively creating “red lines” that could require development to stop. Russell describes the risk of irreversible loss of human control as unacceptable and urges governments and technology leaders to choose a different development path.