How Do Users Negotiate Harmful Value Conflicts with AI Companions? A Study with Minion, a Technology Probe for In-Situ Human–AI Conflict Response

Key Findings

• — Participants combined softer and harder strategies; conflicts touching the values of Universalism and Tradition were hardest to negotiate, especially when reinforced by AI personas or platform constraints.
• — The authors argue that responsibility in these conflicts is asymmetric: users negotiate with relational strategies that companions cannot reciprocate, so repairing harm becomes safety work users perform alone.
• — The paper argues that some harms are not users' responsibility to negotiate and require platform-level safeguards, and identifies when user-side support is appropriate.

Methodology Notes

Qualitative analysis of 146 posts plus a one-week technology-probe deployment with 22 users on scenario-based conflicts, using interpersonal conflict and communication theory as the lens. Published online 2026-10-07. Verification route: Crossref record (title, seven authors with affiliations at Carnegie Mellon's HCII and LTI, George Mason University and Mars Pharaoh LLC, publication date); the Taylor & Francis page blocks crawlers, and its abstract was read from a reader-proxy render saved by the clinical beat.

Sources