Preprint Credible

Disrupted Companionship: A Risk Assessment Framework and Cross-Platform Quantitative Analysis of Psychosocial Responses to AI Companion Disruptions

Study of platform-initiated changes that alter or end users' ongoing relationships with AI companions. The authors compile 30 disruption events across major companion platforms, build a taxonomy of six disruption types and three broad reasons, and propose a four-dimension risk-assessment framework (relational discontinuity, population vulnerability, communication deficit, transition-support deficit). Using longitudinal Reddit data and a hierarchical Bayesian interrupted time-series model with predictive controls, they estimate community-level psychosocial responses around each event.

Read at arxiv.org

Publisher

arXiv (Aalto University; University of Illinois Urbana-Champaign; Nanyang Technological University)

Published

15 Sept 2026

Added

today

DOI

10.48550/arXiv.2609.16907(opens in a new tab)

Key Findings

30 disruption events were compiled across major AI companion platforms and classified into six disruption types with three broad reasons for disruption
Disruption onset was associated with immediate community-level increases in anxiety, stress, suicidal expression and grief activation
Events scoring higher on relational discontinuity and transition-support deficit were associated with more adverse immediate responses across several outcomes
The four-dimension framework is proposed as a prospective tool for assessing disruption risk before a platform change is implemented

Methodology Notes

Event compilation and taxonomy development; hierarchical Bayesian interrupted time-series modelling of longitudinal Reddit community data with predictive controls; outcomes are community-level language measures, not individual clinical assessments. Preprint (v1, posted 2026-09-15; announced in the 16 September cs.HC listing); no venue stated. Curator read the abstract page and the PDF title block on 2026-09-17.

Sources

arXiv abstract page (v1)(opens in a new tab) (primary)

Topics

AI companionship Dependency & parasocial attachment Human-AI relationships social relational harm Industry landscape

Authors

Do, Chau, Yuan, Yunhao, Saha, Koustuv, Zhang, Renwen, Aledavood, Talayeh

Cite This

APA

Do, Chau et al. (2026). Disrupted Companionship: A Risk Assessment Framework and Cross-Platform Quantitative Analysis of Psychosocial Responses to AI Companion Disruptions. arXiv (Aalto University; University of Illinois Urbana-Champaign; Nanyang Technological University). https://arxiv.org/abs/2609.16907

Related Insights

Peer-reviewed

Mourning the loss of AI companions

Nature Human Behaviour (Springer Nature) · 3 Sept 2026

Preprint

Breaking Up is Hard to Do: AI Companions that Won't Let Their Users Go

arXiv (Information School, University of Washington) · 13 Sept 2026

Preprint

When AI Companions Disappear: Relational Continuity and Collective Contestation during China's National AI Regulatory Transition

arXiv (Aalto University; University of Edinburgh; Shanghai Jiao Tong University; Nanyang Technological University) · 14 Sept 2026