Gender and Sexuality Bias in Large Language Models: A Longstanding Issue at an Expanded Scale
-
By
-
Chiara Barbati
-
Virginia Casigliani
-
Caterina Rizzo
-
Anna Odone
-
September 24, 2026
Clinical Report: Gender and Sexuality Bias in Large Language Models
Background
As patients increasingly rely on LLMs for health information, the potential for demographic bias in these systems raises significant public health concerns. Gender bias has been consistently documented and intersects with other demographic factors, potentially exacerbating existing health inequities.
Data Highlights
No numerical or trial data was provided in the source material.
Key Findings
- Gender bias is the most consistently documented form of bias in LLMs, appearing in 15 of 16 studies.
- LLMs can exhibit various biases, including racial, ethnic, age, and socioeconomic biases, alongside gender bias.
- Clinical documentation may reflect gender bias, with male patients receiving more complex wording compared to female patients.
- OpenAI GPT-4 has been shown to perpetuate demographic stereotypes in clinical reasoning tasks.
- Bias in LLM outputs can lead to divergent management recommendations based solely on sociodemographic descriptors.
- Gender bias in LLMs is bidirectional and context-dependent, affecting both men and women.
Clinical Implications
Healthcare professionals should be aware of the potential for biased outputs from LLMs, which may impact clinical decision-making and patient care.
Conclusion
Addressing gender and sexuality bias in LLMs is critical.
Related Resources & Content
- Journal of Medical Internet Research, 2026 -- Evaluating the Potential of Reasoning Large Language Models to Perpetuate Racial and Gender Disease Stereotypes in Health Care
- npj Digital Medicine, 2025 -- The evaluation illusion of large language models in medicine
- Journal of Medical Internet Research, 2026 -- Health Care Professionals' Perspectives on the Integration and Regulation of Large Language Models: A Cross-Sectional Survey Analysis
- Frontiers in Psychiatry — Gender differences in autism prevalence: origins of bias and its current scientific relevance
- HTI-1 Final Rule - ONC - Office of the National Coordinator for Health Information Technology
- Considerations for Generative AI in Public Health | Artificial Intelligence | CDC
- Consensus framework for the validation of generative AI: call for collaborators on the Validation Accords | Nature Medicine
- AMA policies to ensure AI supports—not replaces—physician judgment | American Medical Association
- Bias Evaluation in Medical Applications of Large Language Models: A Systematic Review - PMC
- Large language models exhibit stigmatizing behaviour in contextual judgements of health conditions | Nature Health
- Large language models provide unsafe answers to patient-posed medical questions - PMC
- Journal of Medical Internet Research - Evaluating the Potential of Reasoning Large Language Models to Perpetuate Racial and Gender Disease Stereotypes in Health Care
- Evaluating anti-LGBTQIA+ medical bias in large language models | PLOS Digital Health
- Auditing Large Language Model–Generated Digital Standardized Patients for Demographic Bias: A Simulation Study with HIV Pre-Exposure Prophylaxis Screening as a Tracer Condition | medRxiv
- Measuring stereotype and deviation biases in large language models | Scientific Reports
- Mitigating Automation Bias in Physician-LLM Diagnostic Reasoning Using Behavioral Nudges: A Randomized Controlled Trial | medRxiv
- Evidence, use cases, and implementation safeguards of large language models in primary care | Communications Medicine
Based on findings from:
Sex and gender bias in large language models: an old problem at a new scale
Chiara Barbati, Virginia Casigliani, Caterina Rizzo, Anna Odone. Bmj Health & Care Informatics, 2026.
https://informatics.bmj.com/content/33/1/e102380
This content is an AI-generated, fully rewritten summary based on a published scholarly article. It does not reproduce the original text and is not a substitute for the original publication. Readers are encouraged to consult the source for full context, data, and methodology.
