How AI Can Screw You: OpenEvidence, AI Scribes, and the HIPAA Checkbox

AI integration in medical practice demands rigorous physician oversight to mitigate significant clinical and legal risks. While AI scribes and search tools offer efficiency, they frequently commit "errors of omission," particularly regarding critical mental health and medication data. Clinicians must treat these systems as "smart interns" rather than authoritative attendings, as they are not FDA-approved for emergency settings and often lack the context to handle complex, high-stakes patient scenarios. Furthermore, signing "Business Associate Agreements" without institutional authorization creates severe legal exposure, as physicians may inadvertently assume liability for data breaches. Dr. Sam Ashoo, an emergency physician and AI researcher, emphasizes that because AI models can provide authoritative yet incorrect information, practitioners must verify all outputs against verified clinical knowledge. Relying on AI without independent validation compromises patient safety and shifts the burden of error entirely onto the clinician.