RECORD
Identifying Harm in Personalized, Generative AI Systems Requires User-Centered Auditing at the Interaction Level
Computers and Society · Submitted 7 Aug 2026
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Consequently, we propose reframing understandings of harm as adaptive, user- and community-centered processes, and outline design directions that shift auditing from retrospective evaluation toward infrastructures that support ongoing articulation of harm in interaction.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Personalized, generative AI systems increasingly adapt their behavior to individual users over time, fundamentally changing model behavior. While existing auditing approaches have been effective at surfacing harms in non-personalized contexts, they often rely on static, simulated evaluations and definitions of harm that aggregate across broad, group categories. In this position paper, we argue that such approaches can fail to capture emergent harms in personalized generative AI systems, where harms surface through interpretations of ongoing interaction and evolve with user history. We identify three presuppositions underlying many harm auditing paradigms: that harms can be (1) specified outside real-world interaction, (2) defined non-pluralistically within groups, and (3) treated as static. One might argue that personalized systems could simply learn definitions of what constitutes harm to individual users through repeated interactions. However, we argue that attempts to surface user harms through deeper personalization risk imposing asymmetric burdens of labor and privacy on marginalized users. Consequently, we propose reframing understandings of harm as adaptive, user- and community-centered processes, and outline design directions that shift auditing from retrospective evaluation toward infrastructures that support ongoing articulation of harm in interaction. Our work highlights the need for auditing and design practices that better reflect the pluralistic and evolving nature of harm understanding in personalized generative AI systems.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Personalized, generative AI systems increasingly adapt their behavior to individual users over time, fundamentally changing model behavior. | background 1.00 | 0.07 | 0.03 | asserted 0.99 |
| s1 | While existing auditing approaches have been effective at surfacing harms in non-personalized contexts, they often rely on static, simulated evaluations and definitions of harm that aggregate across broad, group categories. | background 0.51 | 0.04 | 0.03 | asserted 0.76 |
| s2 | In this position paper, we argue that such approaches can fail to capture emergent harms in personalized generative AI systems, where harms surface through interpretations of ongoing interaction and evolve with user history. | gap 0.85 | 0.14 | 0.25 | hedged possibility 0.59 |
| s3 | We identify three presuppositions underlying many harm auditing paradigms: that harms can be (1) specified outside real-world interaction, (2) defined non-pluralistically within groups, and (3) treated as static. | contribution 0.84 | 0.56 | 0.07 | asserted 0.98 |
| s4 | One might argue that personalized systems could simply learn definitions of what constitutes harm to individual users through repeated interactions. | background 0.76 | 0.04 | 0.02 | speculative 0.81 |
| s5 | However, we argue that attempts to surface user harms through deeper personalization risk imposing asymmetric burdens of labor and privacy on marginalized users. | interpretation 0.66 | 0.08 | 0.08 | asserted 0.60 |
| s6 | Consequently, we propose reframing understandings of harm as adaptive, user- and community-centered processes, and outline design directions that shift auditing from retrospective evaluation toward infrastructures that support ongoing articulation of harm in interaction. | contribution 0.84 | 0.56 | 0.85 | asserted 0.98 |
| s7 | Our work highlights the need for auditing and design practices that better reflect the pluralistic and evolving nature of harm understanding in personalized generative AI systems. | interpretation 0.54 | 0.45 | 0.27 | asserted 0.90 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Personalized, generative AI systems increasingly adapt their behavior to individual users over time, fundamentally changing model behavior. While existing auditing approaches have been effective at surfacing harms in non-personalized contexts, they often rely on static, simulated evaluations and definitions of harm that aggregate across broad, group categories. In this position paper, we argue that such approaches can fail to capture emergent harms in personalized generative AI systems, where harms surface through interpretations of ongoing interaction and evolve with user history. We identify three presuppositions underlying many harm auditing paradigms: that harms can be (1) specified outside real-world interaction, (2) defined non-pluralistically within groups, and (3) treated as static. One might argue that personalized systems could simply learn definitions of what constitutes harm to individual users through repeated interactions. However, we argue that attempts to surface user harms through deeper personalization risk imposing asymmetric burdens of labor and privacy on marginalized users. Consequently, we propose reframing understandings of harm as adaptive, user- and community-centered processes, and outline design directions that shift auditing from retrospective evaluation toward infrastructures that support ongoing articulation of harm in interaction. Our work highlights the need for auditing and design practices that better reflect the pluralistic and evolving nature of harm understanding in personalized generative AI systems.
Human review
References
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 30 sentences without a confident label
Builds on 2 papers
-
Prior work has highlighted the challenges of pluralistic alignment, where generative AI systems overlook human social diversity between groups [ref] ; we extend this idea to within-group diversity and highlight the pitfalls of documenting harms and preferences at the group level.
A Roadmap to Pluralistic Alignment · Auditing Approaches Presuppose Non-Pluralistic Definitions of Harm · 0.99
-
Prior work has highlighted the challenges of pluralistic alignment, where generative AI systems overlook human social diversity between groups [ref] ; we extend this idea to within-group diversity and highlight the pitfalls of documenting harms and preferences at the group level.
2511.14476 · Auditing Approaches Presuppose Non-Pluralistic Definitions of Harm · 0.99
Cites as support 7 papers
-
Users can discover nuanced and community-specific harms that have previously been overlooked by researchers or practitioners [ref] .
2503.19075 · Surfacing Harms Via Algorithm Auditing · 1.00
-
Additionally, even if users were to articulate harms as they occur, the lack of transparency in current AI systems [ref] and the difficulty for users to steer AI system behavior [ref] make it unclear whether this labor would prevent harms in future outputs.
2503.17482 · Asymmetry of Labor · 0.99
-
Additionally, even if users were to articulate harms as they occur, the lack of transparency in current AI systems [ref] and the difficulty for users to steer AI system behavior [ref] make it unclear whether this labor would prevent harms in future outputs.
AI data transparency: an exploration through the lens of AI incidents · Asymmetry of Labor · 0.99
Cites as background 3 papers
-
Recent work has attempted to better understand and evaluate how personalized interaction can affect behaviors for generative AI systems, such as chatbots [ref] .
2511.01805 · Personalized, Generative AI Systems · 0.82
-
Algorithm auditing describes the process of surfacing harms by directly analyzing the outputs of such systems [ref] .
AI auditing: The Broken Bus on the Road to AI Accountability · Surfacing Harms Via Algorithm Auditing · 0.86
-
Generative AI systems are being increasingly integrated into everyday applications, such as chatbots [ref] and writing assistants [ref] .
A Complete Survey on LLM-based AI Chatbots · Introduction · 0.96
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.