Overview
- Anthropic published its proposal on Friday calling for a coordinated global slowdown in the most advanced AI work and released internal analyses showing models increasingly doing development tasks that humans once handled.
- The company warned of recursive self‑improvement, meaning systems could help design or refine their own successors, and said that while that stage has not been reached it appears closer than expected.
- Anthropic said an effective pause would need participation from major firms and governments, independent verification of compliance, and clear rules, and it plans to convene officials, academics, advocacy groups and rival companies to explore how such a mechanism could work.
- The proposal has drawn swift pushback from industry and some officials who say a unilateral or uncoordinated pause could cede advantage to rivals, and the White House this week issued an executive order that allows reviews of powerful U.S. models before public release.
- Experts and Anthropic note major obstacles to any pause, including hard‑to‑detect model training, strong commercial incentives, U.S.–China strategic competition, and the short‑term economic pressures around IPOs and investment that keep firms racing to deploy new capabilities.