Frontier AI leaders’ stated stance shifts toward slower development and outside oversight
PolicyUnverifiedUpdated:
Frontier-AI leaders reportedly backed slower development and outside safety oversight. The report describes a change in public posture among companies that had competed to build more capable models.
Slower development and outside evaluators inside labs
Ars Technica reports that Amodei urged AI developers to slow capability gains because of potential risks from increasingly capable systems. The article says other industry leaders supported deliberate pacing, common standards, or safety measures. It also describes a proposed model in which independent organizations place evaluators inside frontier-AI labs with employee-like access to assess safety practices and report incidents.
According to the report, major frontier-AI labs had for years operated in a winner-take-all race to improve model capabilities.
Anthropic’s Dario Amodei publicly called for slower progress in frontier AI, and leaders at OpenAI, Google DeepMind, Microsoft, and others echoed safety-focused pacing or coordination. Anthropic said it would use outside embedded evaluators, while OpenAI’s Sam Altman said OpenAI would do the same.
If these stated commitments are carried out, the change would move the debate from general calls for AI safety toward potential outside monitoring and coordinated limits on development speed. The report also notes that the proposals remain uncertain and could align with commercial interests as well as safety goals.
Sources
- Ars TechnicaAI leaders want to hit the brakes after years of reckless speedRead source →