Anthropic research shows AI agents are “self-improving” without human guidance

Anthropic researchers have demonstrated an automated system capable of developing techniques to reduce problematic behaviour in artificial intelligence models.

Published: 2026-08-28