
Behavioral trajectories are becoming increasingly crucial in agentic AI safety and we need to understand how they evolve. Excited to share that our paper, “Revealing Echoes of Human Malice: Benchmarking LLM Agents for Multi-Turn Online Harassment Attacks” has been accepted to AAAI-ICWSM 2027. Preprint: arxiv.org/pdf/2510.14207 @icwsm 🧵 1/8






































