AI RESEARCH
Multi-Trait Subspace Steering to Reveal the Dark Side of Human-AI Interaction
arXiv CS.AI
•
ArXi:2603.18085v1 Announce Type: new Recent incidents have highlighted alarming cases where human-AI interactions led to negative psychological outcomes, including mental health crises and even user harm. As LLMs serve as sources of guidance, emotional, and even informal therapy, these risks are poised to escalate.