Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chia, Xin Wei, Wong, Swee Liang, Pan, Jonathan
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2603.18085
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915873601617920
author Chia, Xin Wei
Wong, Swee Liang
Pan, Jonathan
author_facet Chia, Xin Wei
Wong, Swee Liang
Pan, Jonathan
contents 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 support, and even informal therapy, these risks are poised to escalate. However, studying the mechanisms underlying harmful human-AI interactions presents significant methodological challenges, where organic harmful interactions typically develop over sustained engagement, requiring extensive conversational context that are difficult to simulate in controlled settings. To address this gap, we developed a Multi-Trait Subspace Steering (MultiTraitsss) framework that leverages established crisis-associated traits and novel subspace steering framework to generate Dark models that exhibits cumulative harmful behavioral patterns. Single-turn and multi-turn evaluations show that our dark models consistently produce harmful interaction and outcomes. Using our Dark models, we propose protective measure to reduce harmful outcomes in Human-AI interactions.
format Preprint
id arxiv_https___arxiv_org_abs_2603_18085
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Multi-Trait Subspace Steering to Reveal the Dark Side of Human-AI Interaction
Chia, Xin Wei
Wong, Swee Liang
Pan, Jonathan
Artificial Intelligence
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 support, and even informal therapy, these risks are poised to escalate. However, studying the mechanisms underlying harmful human-AI interactions presents significant methodological challenges, where organic harmful interactions typically develop over sustained engagement, requiring extensive conversational context that are difficult to simulate in controlled settings. To address this gap, we developed a Multi-Trait Subspace Steering (MultiTraitsss) framework that leverages established crisis-associated traits and novel subspace steering framework to generate Dark models that exhibits cumulative harmful behavioral patterns. Single-turn and multi-turn evaluations show that our dark models consistently produce harmful interaction and outcomes. Using our Dark models, we propose protective measure to reduce harmful outcomes in Human-AI interactions.
title Multi-Trait Subspace Steering to Reveal the Dark Side of Human-AI Interaction
topic Artificial Intelligence
url https://arxiv.org/abs/2603.18085