TY  - RPRT
TI  - CompanionHarm: A Multi-Turn Benchmark for Detecting Harms in Real-World AI Companion Conversations
AU  - Renwen Zhang
AU  - Han Meng
AU  - Jian Chai
AU  - Yuntao Lin
AU  - Yi-Chieh Lee
PY  - 2026
UR  - https://arxiv.org/abs/2608.25377
ID  - 2608.25377
ER  -