When Single-User-Oriented LLM-based Assistants Involve Others: A Scoping Review of Pathways, Risks, and Responses
A scoping review of 58 studies on how LLM-based assistants built around a single user come to involve other people, given that context management, personalisation, identity attribution, authority attribution and action execution are organised around one user. It identifies five operational pathways of direct and indirect involvement, five recurring risk domains for the additional parties, and five areas of implemented or proposed responses, and argues for governance that follows changing cross-person roles and for designs that preserve person-specific boundaries throughout interaction.
Publisher
arXiv (North Carolina State University; University of Hong Kong; Carnegie Mellon University alumnus)
Published
12 Sept 2026
Added
today
Key Findings
- 58 studies were reviewed on single-user-oriented assistants extending into multi-party contexts.
- Five operational pathways were identified through which an assistant involves additional human parties, spanning direct and indirect involvement.
- Five recurring risk domains and five areas of implemented or proposed responses were mapped.
- The authors call for governance attentive to changing cross-person roles and relationships and for assistant designs that preserve person-specific boundaries.
Methodology Notes
Scoping review (58 studies), 18 pages; the pathway and risk labels are in the paper body. arXiv v1 submitted 2026-09-12 and readable at the abstract page, not yet announced in a daily listing at the time of entry. Affiliations from the PDF title block. No empirical data of its own.
Sources
Topics
Authors
Yulin Chen, Yang Zhan, Zhuoran Lu, Qiao Jin
Tags
Cite This
APA
Yulin Chen et al. (2026). When Single-User-Oriented LLM-based Assistants Involve Others: A Scoping Review of Pathways, Risks, and Responses. arXiv (North Carolina State University; University of Hong Kong; Carnegie Mellon University alumnus). https://arxiv.org/abs/2609.14062