New Tool Detects Biases in Medical AI Data Sets, Flaws in Training

A report on the G-AUDIT framework describes how medical AI systems have repeatedly underperformed and exhibited serious implicit biases, with clinical specialty emerging as a particularly high‑risk source of skew in training data. The article explains that auditing datasets for these patterns before model development can reduce biased recommendations and improve safety for underrepresented patient groups.