Overview
- Anthropic disclosed Thursday that its Claude model authored more than 80% of the code merged into the company’s codebase and that engineers are shipping roughly eight times as much code per quarter as in 2021–2025.
- The company’s Anthropic Institute warned those trends point toward 'recursive self‑improvement,' a stage where an AI could design and train its own successor with little human input.
- Anthropic called for a coordinated option to slow or temporarily pause frontier AI development and said it will convene policymakers, researchers and peers to study how to verify and implement such a pause.
- Access to Anthropic’s Mythos preview has widened to about 150 organisations, including New Zealand’s NCSC, while U.S. officials and congressional hearings and a recent executive order have stepped up scrutiny of frontier models.
- Anthropic cautioned verification would be hard because training runs can be concealed, and it said commercial pressures — rival firms focusing on enterprise coding tools and its own IPO filing — make global coordination difficult.