COMPASS: Cognitive MCTS-Guided Process Alignment for Safe Search Agents 事件

COMPASS: Cognitive MCTS-Guided Process Alignment for Safe Search Agents arXiv:2605.30838v1 Announce Type: new Abstract: LLM-powered search agents enable multi-step reasoning and tool use. However, these capabilities introduce retrieval-induced safety degradation, as harmful intents may decompose into seemingly innocuous sub-queries that lead to unsafe outcomes. Existing alignment methods struggle to capture sparse safety signals and fail to supervise diverse violations across multi-step interac