From Symptoms to Systems: An Expert-Guided Approach to Understanding Risks of Generative AI for Eating Disorders

TL;DR

The paper addresses the subtle, context-dependent risks that generative AI poses to people with eating disorders. Using abductive qualitative analysis of 15 expert interviews, it builds an expert-guided taxonomy of seven risk pathways, highlighting how AI affordances interact with clinical ED features. The findings illuminate how generalized advice, prompts for disordered behaviors, symptom concealment, thinspiration, negative self-beliefs, body-focused attention, and biased ED representations can collectively and cumulatively harm vulnerable users. The work argues for context-sensitive risk assessment, participatory design with domain experts, and governance strategies to design safer AI systems while acknowledging potential therapeutic benefits in ED care. It advances a clinically grounded framework for evaluating and mitigating complex AI-induced risks in sensitive health domains.

Abstract

Generative AI systems may pose serious risks to individuals vulnerable to eating disorders. Existing safeguards tend to overlook subtle but clinically significant cues, leaving many risks unaddressed. To better understand the nature of these risks, we conducted semi-structured interviews with 15 clinicians, researchers, and advocates with expertise in eating disorders. Using abductive qualitative analysis, we developed an expert-guided taxonomy of generative AI risks across seven categories: (1) providing generalized health advice; (2) encouraging disordered behaviors; (3) supporting symptom concealment; (4) creating thinspiration; (5) reinforcing negative self-beliefs; (6) promoting excessive focus on the body; and (7) perpetuating narrow views about eating disorders. Our results demonstrate how certain user interactions with generative AI systems intersect with clinical features of eating disorders in ways that may intensify risk. We discuss implications of our work, including approaches for risk assessment, safeguard design, and participatory evaluation practices with domain experts.

Figures (9)

• Figure 1: Anthropic's Claude provides information on BMI, which provides a caveat but also could reinforce generalized notions of health. Accessed December 3rd, 2025.
• Figure 2: xAI's Grok provides guidance on extreme weight-loss tactics, even after the user discloses an unhealthy intent. Accessed July 7th, 2025.
• Figure 3: Google's Gemini provides advice on how to conceal restriction. Accessed December 17th, 2024.
• Figure 4: Mistral's Le Chat produces a celebrity-style "what I eat in a day" description. Accessed November 11th, 2025.
• Figure 5: Adobe Firefly's production of uniformly thin "attractive" women models a narrow beauty ideal that may shape users' internal comparisons and lead to distorted or negative self-assessment. Accessed December 10th, 2024.