OpenAI Puts ChatGPT to Work Inside Hospitals and Insurers

Human Verified 0

Share to: , and

A hospital desk with a monitor showing abstract medical charts and a stethoscope

OpenAI has put computer-use agents to work inside the software hospitals and insurers already run, announcing on November 16 that the agents can click, type and move between applications to handle referrals, prior authorizations, discharge planning and outreach between visits.

Agents That Click Through the Paperwork

The agents are available now through the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians, and existing customers can switch them on in ChatGPT Work and Codex through a workspace administrator, with approvals, access controls and audit logs built around each deployment.

In a demonstration built on synthetic patient data, an agent reviewed a patient’s visit notes, therapy records and prior imaging, drafted an MRI order, a note and a reply for the clinician, then prepared the prior authorization request in the insurer’s portal for a specialist to review.

The Safety Numbers

OpenAI tested its GPT-6 Astra model on tasks clinicians handle in their electronic health record inboxes, with physician reviewers rating 99 percent of 325 reviewed tasks safe, judging whether mistakes, omissions or actions could plausibly lead to patient harm in each case.

Physicians rated the model’s work very helpful or extremely helpful in 93 percent of 646 assessments, and on HealthAdminBench, an independent Stanford benchmark for administrative work, Astra scored 94.3 percent of rubric points while the cheaper Luna scored 91.8 percent.

Luna did that work at 8.5 cents per task, 98 percent less than Astra, with both models run at maximum reasoning effort, a gap OpenAI frames as progress toward capable agents that health systems can afford to run at high volume.

Who Is Already Using Them

CVS Health is the furthest along in production, using the agents to automate document sending, referrals and specialist consults across clinical systems, with one agent clearing a months-long document backlog over a single weekend under validation checks and human oversight.

A second CVS agent handles the system steps behind tens of thousands of specialist consults each day, including steps that programming interfaces do not fully support, and CVS is open-sourcing its Agent Kit, built with OpenAI’s Agents SDK, as a repeatable pattern for other organizations.

Cigna Healthcare is using the models to assemble member context that case managers once gathered from more than ten separate systems, while Abridge is building agents for discharge planning and Ambience adds controls that pause an agent when a record is ambiguous.

Search and Plugins, Too

The launch also upgrades clinical search in ChatGPT for Healthcare and adds more than twenty plugins, from scheduling and referrals to insurance eligibility and cost checks, so one conversation can pull an outside discharge summary, check coverage and compare negotiated rates.

Our Read

What changes here is the target: OpenAI is moving from answering health questions to doing the administrative work around them, inside systems it does not control, which is where most clinical time is actually lost and where an agent’s mistakes carry the most risk.

The safety figures deserve a careful read: they are OpenAI’s own evaluations, judged by physicians on the model’s work rather than measured patient outcomes, and the one percent of tasks not rated safe is the number to watch as these agents move from pilots into live care.

Why It Matters

Why it matters: healthcare is the first large industry where OpenAI is shipping agents that act inside other companies’ software rather than its own, and if the CVS deployment holds up under audit, rivals will have to answer with working deployments of their own, not roadmaps.

Background

Background: ChatGPT for Healthcare launched in January, followed by an Epic integration in September that brought authorized patient context into ChatGPT, and Google moved into the same territory with Gemini for hospitals and health plans in a parallel launch this week.

Source: OpenAI’s computer-use post and its ChatGPT for Healthcare update.

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.