This quality improvement study evaluates the susceptibility of commercial large language models (LLMs) to prompt-injection attacks that may induce unsafe clinical advice contraindicated treatments.