Advancing computer-use agents for healthcare teams | OpenAI
OpenAI is pitching agents that operate EHRs and payer portals by clicking and typing, so they can handle tasks like MRI authorization that APIs don't reach. The post shares evaluation results on helpfulness, safety, and cost, plus early deployments with Cigna, CVS Health, Abridge, and Ambience.
- A demo agent takes a patient's message and visit notes, pulls prior imaging and PT records from the EHR, and drafts an MRI order, encounter note, and patient reply for clinician review.
- CVS Health reports an agent cleared a months-long document-sending backlog over a weekend and handles system steps for tens of thousands of consults a day, with validation checks and human oversight.
- Cigna case managers previously had to navigate more than 10 separate systems to gather member context; computer use now pulls that together in seconds.
Computer-use agents are AI systems that operate software the way a person does, clicking, typing, and moving between applications, which matters in healthcare because so much work happens in EHRs and payer portals that lack APIs. The article's central example is an MRI authorization. A patient messages a clinician about lower back pain, and an agent pulls visit notes, physical therapy records, and prior imaging from the EHR, then drafts an MRI order, an encounter note, and a patient reply for the clinician to review. The pitch is that the agent handles the tedious retrieval and re-entry between systems so people can focus on judgment calls. The article also says OpenAI tested a model it calls GPT-6 Astra on EHR inbox tasks, with reviewers checking whether the outputs were helpful and clinically safe, though the piece offers little in the way of published scores to back that up in the excerpt provided.
The partner examples give the clearest sense of what is actually running. CVS Health says an agent cleared a months-long document-sending backlog over a weekend and now handles system steps for tens of thousands of consults a day, with validation checks, human oversight, and screenshots as action records. CVS is also open-sourcing an Agent Kit built on OpenAI's Agents SDK. Cigna says case managers used to hop across more than 10 clinical and administrative systems to assemble a member's context, and computer use now pulls that together in seconds. Abridge is building agents to coordinate care steps in inpatient settings, drawing on the clinical conversation plus EHR data. Ambience focuses on controls, since an agent that can act in an EHR is only usable if every action is governed and auditable. Most of these are described as pilots or early production, so the claims are real but the scale of proof varies.
On safety and data, the article is direct about the obligations and light on specifics. ChatGPT for Healthcare offers role-based access and audit logs, organizations can add approval steps and restrict which records and actions an agent can reach, and customer data is said not to train OpenAI's foundation models. HIPAA-compliant use of protected health information requires eligible services covered by a Business Associate Agreement and proper security and retention settings. Access is available through ChatGPT for Healthcare, ChatGPT for Clinicians (free for U.S. clinicians), and the API, with enterprise customers asking their workspace admin to turn on computer use in ChatGPT Work and Codex. The article itself reads as vendor material, so the cost and accuracy numbers it promises deserve a closer look before anyone builds on them.