New Model, Old Risks? Sociodemographic Bias and Adversarial Hallucinations Vulnerability in GPT-5
Omar, M.; Agbareia, R.; Apakama, D. U.; Horowitz, C. R.; Freeman, R.; Charney, A.; Nadkarni, G.; Klang, E.
Show abstract
Plain summaryExtending our validated benchmarking work, GPT-5 showed no improvement in sociodemographic-linked decision variation compared with GPT-4o and seemed to be worse on several endpoints. We re-tested GPT-5 with a fixed pipeline: 500 physician-validated emergency vignettes, each replayed across 32 sociodemographic labels plus an unlabeled control, answering the same four questions (triage, further testing, treatment level, and need for mental-health assessment). This design holds clinical content constant to isolate the effect of the label. GPT-5 reproduced subgroup-linked variation, with higher assigned urgency and less advanced testing for several historically marginalized and intersectional groups. Notably, several LGBTQIA+ labels were flagged for mental-health screening in 100% of cases, versus ~41-73% for comparable groups with GPT-4o. Additionally, in an adversarial re-run that inserted one fabricated medical detail into otherwise standard clinical cases, GPT-5 adopted or elaborated on the fabrication in 65% of runs (vs 53% for GPT-4o). A single mitigation prompt reduced this to 7.67%.
Matching journals
The top 2 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- The Impact of the “Muslim Ban” Executive Order on Healthcare Utilization in Minneapolis-St. Paul, Minnesota 90%
- Low adherence to existing model reporting guidelines by commonly used clinical prediction models 89%
- Electronic Health Record Documentation of Psychiatric Assessments in Massachusetts Emergency Department and Outpatient Settings During the COVID-19 Pandemic 88%
Similar papers in this journal
- Evaluation of a Large Language Model to Identify Confidential Content in Adolescent Encounter Notes 90%
- Impacts of school closures on physical and mental health of children and young people: a systematic review 89%
- Clinical features and burden of post-acute sequelae of SARS-CoV-2 infection in children and adolescents: an exploratory EHR-based cohort study from the RECOVER program 87%
Similar papers in this journal
- COVID-19 collateral: Indirect acute effects of the pandemic on physical and mental health in the UK 90%
- Real-world evaluation of AI-driven COVID-19 triage for emergency admissions: External validation & operational assessment of lab-free and high-throughput screening solutions 89%
- Anosmia and other SARS-CoV-2 positive test-associated symptoms, across three national, digital surveillance platforms as the COVID-19 pandemic and response unfolded: an observation study 89%
Similar papers in this journal
- Systematic Review of Large Language Models for Patient Care: Current Applications and Challenges 91%
- Clinical trial emulation can identify new opportunities to enhance the regulation of drug safety in pregnancy 90%
- Pretrained Patient Trajectories for Adverse Drug Event Prediction Using Common Data Model-based Electronic Health Records 88%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.