A recent arXiv paper argues that autonomous AI systems should have three key features to safely prescribe medications: calibrated per-prediction confidence, differentiated communication of uncertainty types, and inferential transparency. The authors surveyed 136 US prescribing clinicians and found that they would not permit autonomous prescribing without a confidence-based escalation mechanism, preferred competing-options summaries for aleatoric uncertainty, and only accepted liability when inferential transparency enabled them to make decisions under acknowledged uncertainty. These findings suggest that incorporating these features could increase clinician adoption of autonomous AI prescribing. --- Why it matters: This matters because it highlights the need for AI systems to be transparent and accountable in high-stakes medical decision-making. Clinicians want to understand how AI recommendations are made and have mechanisms in place to handle uncertainty, which is crucial for safe autonomous prescribing. Source: https://arxiv.org/abs/2606.25108