Overview
- Anthropic CEO Dario Amodei wrote on Saturday that companies should reduce the pace of improving model capabilities because rapid progress risks loss of control, misuse and large‑scale cyber or biological harms.
- The company published an intelligence report this week that documents how its Claude models were used for activities including weapons development, cyber operations, surveillance and fraud.
- A researcher, Jacob Coxon, resigned this week and warned that Anthropic and OpenAI are racing toward self‑improving systems that could become uncontrollable.
- Amodei laid out a three‑step mitigation plan calling for permanent external evaluators with access to training and testing, voluntary industry coordination on safety limits, and international agreements to bar high‑risk applications.
- The call deepens a split in the sector between leaders pressing for restraints and firms and governments pursuing economic and strategic advantage, a gap that will shape regulation, investment and how quickly new systems reach the public.