Agents in the clinic

OpenAI is expanding its computer-use technology into healthcare. The AI agents navigate clinical software through clicks and keyboard input, retrieve patient information, prepare medical documentation and coordinate insurance authorization requests across separate applications. The goal is to cut the time US healthcare staff spend moving information between disconnected systems.

How it works

In an OpenAI demonstration, an agent processes an MRI request for a patient with lower back pain. It first reviews the electronic health record (EHR), including prior consultations, physical therapy records and imaging history. It then drafts an MRI order, a clinical note and a patient response, and uses the same records to prepare a prior authorization request through an insurer's portal. Clinicians and authorization specialists review the documents before any further action.

Partners and field results

OpenAI is working with CVS Health, Cigna Healthcare, Abridge and Ambience. CVS Health uses agents to send documents, process referrals and support specialist consultations; one agent cleared a document-sending backlog that had built up over several months in a single weekend. Cigna Healthcare uses the technology to help case managers gather information across more than 10 separate systems. Abridge is developing agents for inpatient care coordination, while Ambience uses computer interaction to prepare insurance authorization requests and pauses for clinician input when records are ambiguous.

Safety and performance

OpenAI evaluated GPT-6 Astra on electronic health record inbox tasks. Physicians rated 99% of 325 reviewed tasks as safe, and 93% of 646 assessments as very or extremely helpful. On Stanford researchers' HealthAdminBench, Astra scored 94.3% of available points and GPT-6 Luna scored 91.8%. OpenAI also reported a substantial cost difference between the two models.