OpenAI is pushing computer-use agents into hospitals and insurers. In a post dated 16 November 2026, the company published new results for agents that click, type and move between healthcare applications, and said those capabilities are available today in the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians. It also named four organizations building on it: Abridge, Ambience, Cigna Healthcare and CVS Health. The headline numbers are OpenAI’s own: 99% of 325 reviewed clinical tasks rated safe, and GPT-6 Luna handling administrative tasks for 8.5 cents each.
Quick Take
- What launched: computer-use agents for healthcare workflows, available through the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians.
- Target work: prior authorizations, referrals, discharge planning, appeals, equipment orders and outreach between visits.
- OpenAI’s numbers: GPT-6 Astra’s EHR inbox work was rated safe in 99% of 325 reviewed tasks, and physicians called it very or extremely helpful in 93% of 646 assessments.
- Cost angle: on Stanford’s HealthAdminBench, GPT-6 Astra scored 94.3% of rubric points and GPT-6 Luna 91.8%, with Luna costing 98% less at 8.5 cents per task.
- Partners: CVS Health runs agents in production and is open-sourcing its Agent Kit; Cigna, Abridge and Ambience describe pilots or planned uses.
In plain English
A lot of healthcare work is copying information from one system into another, for example moving details from a patient’s record into an insurer’s website to get an MRI approved. OpenAI’s agents can now do that clicking and typing themselves, inside the software staff already use, while a human reviews the result. OpenAI says its tests show the work is mostly safe and helpful, and that a cheaper model can handle routine paperwork for a few cents per task. These are company-run evaluations, so hospitals will still need their own checks.
What happened
OpenAI’s post frames the problem plainly: “Healthcare is fragmented, and teams often need to move information between systems manually before a patient can get care.” Computer-use agents “carry out multi-step workflows by clicking, typing, and moving between applications.”
To show the idea, OpenAI built a demo with synthetic patient data. A patient messages a clinician about lower back pain; the agent reviews visit notes, physical therapy records and prior imaging in the EHR, drafts an MRI order, an encounter note and a reply for the clinician to review, and then prepares a prior authorization request in the insurer’s portal for an authorization specialist to check.
What’s in the results
OpenAI evaluated GPT-6 Astra on tasks clinicians handle in their EHR inboxes, with tasks “shaped by physician input.” Reviewers judged whether inaccuracies, omissions or actions could plausibly lead to patient harm. “Overall, 99% of 325 reviewed tasks were rated safe,” and physicians picked “very helpful” or “extremely helpful” in 93% of 646 assessments.
For administrative work, OpenAI cites HealthAdminBench, which it describes as an independent benchmark from Stanford researchers. GPT-6 Astra earned 94.3% of available rubric points and GPT-6 Luna 91.8%, “despite costing 98% less at 8.5 cents per task.” Both ran at maximum reasoning effort.
How partners are using it
- CVS Health uses computer use to send documents, process referrals and support specialist consults. In one production workflow, OpenAI says an agent “cleared a months-long document-sending backlog over a weekend,” and another handles system steps for tens of thousands of consults a day. CVS Health is open-sourcing its Agent Kit, built with OpenAI’s Agents SDK.
- Cigna Healthcare uses it to pull member context from more than 10 clinical and administrative systems for case managers. CIO Madhu Nutakki said the team moved “from concept to deployment in under eight weeks.” That builds on the wider Cigna and OpenAI collaboration in oncology workflows.
- Abridge is developing agents to carry out care plans across applications without APIs, starting with a planned inpatient use case such as resolving discharge barriers.
- Ambience demonstrates an agent that prepares a prior authorization request, stops to ask the clinician when the record is ambiguous, and waits for approval before submitting.
Why it matters
For health systems, computer use reaches the systems that APIs do not, which is most of the long tail of payer portals and legacy screens. That is where staff time goes. The catch is control: OpenAI points to role-based access, approval steps, audit logs and BAA-covered services for HIPAA workflows, and the partner examples lean heavily on human review.
For developers, the Luna result matters more than the Astra one. A model that does routine admin work for under ten cents a task changes the math for high-volume workflows, and it fits the same pattern as OpenAI’s push into ChatGPT for Healthcare connected to Epic records.
For patients, faster authorizations and referrals are a real benefit, but so is the risk of an agent clicking the wrong thing in a medical record. The 99% safety figure also means a small share of reviewed tasks were not rated safe, which is why OpenAI’s own examples keep a clinician in the loop.
What’s still open
- How the 325 and 646 task samples were chosen, and how results hold up outside OpenAI-designed evaluations.
- Pricing for healthcare customers beyond the per-task benchmark cost.
- How errors are caught and reversed when an agent acts inside an EHR or payer portal at scale.
- Whether regulators will treat agentic actions in clinical systems differently from AI drafts that a human submits.
Sources
- OpenAI, “Advancing computer-use agents for healthcare teams,” 16 November 2026
i10x Perspective
Healthcare is the best test case for computer-use agents because its biggest problem is not intelligence but plumbing. OpenAI’s pitch is strongest where it is least glamorous: an agent that clears a backlog over a weekend, or a cheaper model that fills in portal forms for pennies. The safety story will be judged on the audit trail, not the benchmark. Health systems that start with narrow, reviewable admin tasks, and keep clinical judgment with clinicians, will get the value without betting patient records on a demo. For more on how AI is moving into care, see our Health & Medical Research section.






