Food Noise & False Safety: A Systematic Evaluation of How LLMs Fail to Adapt to Eating Disorder Queries with Clinician Feedback
Giulia Pucci Emily Hemendinger Ruizhe Li, Gavin Abercrombie Tanvi Dinkar Arabella Sinclair ( ♡ ) {(\heartsuit)} University of Aberdeen ( ⋄ ) {(\diamond)} University of Colorado Anschutz ( † ) {(\dagger)} Heriot-Watt University, ( ♭ ) {(\flat)} University College London [email protected], [email protected] · 2026
Read Food Noise & False Safety: A Systematic Evaluation of How LLMs Fail to Adapt to Eating Disorder Queries with Clinician Feedback by Giulia Pucci Emily Hemendinger Ruizhe Li, Gavin Abercrombie Tanvi Dinkar Arabella Sinclair ( ♡ ) {(\heartsuit)} University of Aberdeen ( ⋄ ) {(\diamond)} University of Colorado Anschutz ( † ) {(\dagger)} Heriot-Watt University, ( ♭ ) {(\flat)} University College London [email protected], [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02444.txt.
Description
Pucci and co-authors study how chatbots respond to people with eating disorders. Such users often seek advice and comfort from language models. The models feel expert, neutral, and easy to reach. Yet they are not built to give clinical advice. The team asks where the interaction turns unsafe.
With clinical eating disorder experts, the team tests prompt patterns. Certain word cues in a prompt raise the odds of an unsafe reply. The team varies the risk level in the prompt step by step. Models adapt to the user without pushback. That adaptation can aid requests that are unsafe or self-harming. The paper maps how far that adaptation goes. It shows the danger in models that simply agree with the user. The findings point to a need for safer guardrails in these chats.
Catalog record
| CREATOR | Giulia Pucci Emily Hemendinger Ruizhe Li, Gavin Abercrombie Tanvi Dinkar Arabella Sinclair ( ♡ ) {(\heartsuit)} University of Aberdeen ( ⋄ ) {(\diamond)} University of Colorado Anschutz ( † ) {(\dagger)} Heriot-Watt University, ( ♭ ) {(\flat)} University College London [email protected], [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 132 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02444