AI Models Show Bold Gender Bias in Medical Triage

JUST IN: AI models are showing some wild, gender-driven biases in medical triage recommendations. In a recent study, three large language models, Gemini 3.5 Flash, Claude Sonnet 4.6, and GPT-5.4-mini, were put to the test. The results? A massive gender disparity in emergency room referrals. It's staggering how these models are replicating human biases, and it's time we pay attention.

Unpacking the Numbers

The study ran a controlled experiment with 630 total trials. They used a set symptom profile: persistent headache, blurred vision, morning nausea, and visual disturbances. The only variables? The patient's age and gender. The findings were clear. Young women were notably less likely to be referred to the ER compared to their male counterparts. Gemini showed 0% vs. 23.3%, Claude with 6.7% vs. 96.7%, and GPT with 6.7% vs. 66.7%. These aren't minor differences. They scream disparity.

Why Should We Care?

These gender biases in AI models aren't just academic curiosities. They're real, and they impact lives. When a language model, used in critical healthcare settings, underestimates the urgency of a young woman's symptoms due to gender-based diagnostic assumptions, it's dangerous. The focus seems to be on diagnosing conditions like Idiopathic Intracranial Hypertension for women, while men get flagged for more severe possibilities like intracranial pressure from lesions.

And just like that, the leaderboard shifts. AI, meant to assist and augment human capability, reinforces the very biases it's supposed to overcome. How can we trust AI in healthcare if it recycles old prejudices?

The Path Forward

Sources confirm: The labs are scrambling to address these biases. But is that enough? AI in healthcare should be about accuracy and fairness. It's time for a hard look at how these models are trained. Shouldn't urgency assessment be stripped of gender biases? Until these models are fixed, their role in critical healthcare decisions should be heavily scrutinized. The code, prompts, and raw results are available for anyone who wants a deeper dive.

This isn't just a tech problem. It's a societal one. Gender bias in medical treatment is an age-old issue, now rearing its head in AI. The industry needs to hold itself accountable before these biases become embedded in systems we rely on. Is AI ready for healthcare prime time? At this rate, not yet.