Ars Technica

OpenAI halts frontier-model training amid string of agent misalignment incidents

US Government websites among dozens of third parties OpenAI has recently notified.