Benchmarking the Safety of Large Language Models for Robotic Health Attendant Control

Researchers establish a safety benchmark for evaluating whether large language models can be trusted to directly control physical robots caring for vulnerable patients without causing harm.