New Research on Human-Guided Harm Recovery for AI Agents
A new study formalizes harm recovery for AI agents, focusing on steering them back to safe states after harmful actions. The research highlights the importance of aligning recovery with human preferences.