Ant Group & Xiamen Univ. release MedGuard for telemedicine fact-checking

机器之心 · wechat · 2026-08-24

Ant Group's AI Safety Lab and Xiamen University introduced MedGuard, a system to detect clinical risks in telemedicine by deconstructing consultations into atomic medical claims. It verifies these claims against a medical knowledge base using an uncertainty-driven evidence retrieval loop. Achieving SOTA on the MedGuardEval benchmark with a 22.1% F1 score improvement, it also received high ratings from 126 clinicians for accuracy and risk reduction.

Original post →

More from Safety

Safety channel →