Independent · Source-cited · No sponsored rankingsSources linked in every article
AIToolRanked
AI News · 5 min read

ChatGPT for Healthcare adds computer-use agents for prior auth

OpenAI added computer-use agents to ChatGPT for Healthcare on November 16, 2026, for work such as prior authorization. U.S. clinicians can sign up free.

RA
· Founder, AIToolRanked
Three overlapping software windows with charts and form fields, and a cursor moving from the center window toward the window on the right
On this page

What OpenAI shipped on November 16

On November 16, 2026, OpenAI published More ways to move care forward with ChatGPT for Healthcare ↗ and Advancing computer-use agents for healthcare teams ↗. HIT Consultant ↗ covered the launch the same day, from the HLTH conference. The picture above is ours. It is not a screenshot of an electronic health record, and it is not OpenAI's social card. This page is not medical advice.

Computer-use agents read screens, click, type, and move between applications. OpenAI's healthcare examples are referrals, prior authorizations, discharge planning, and outreach between visits, including electronic health records and payer portals. The capabilities are available today in ChatGPT for Healthcare, ChatGPT for Clinicians, and the OpenAI API. Existing customers ask a workspace administrator to enable computer use in ChatGPT Work and Codex. U.S. clinicians can sign up for a free ChatGPT for Clinicians account.

Controls on the posts: role-based access, audit logs, and configurable approval steps and app access. HIPAA workflows that touch protected health information require services covered by a business associate agreement with OpenAI, plus the required security and retention settings. The computer-use post says organization data is not used to train OpenAI foundation models. The healthcare post says content shared with ChatGPT for Healthcare is not used to train models.

What OpenAI measured, and who is using it

The scores below are OpenAI's. This desk did not run the tasks.

On electronic-health-record inbox tasks, OpenAI says reviewers rated GPT-6 Astra clinically safe on 99% of 325 reviewed tasks. Physicians picked "very" or "extremely" helpful in 93% of 646 assessments. On HealthAdminBench, which OpenAI calls an independent benchmark from Stanford researchers, OpenAI says GPT-6 Astra earned 94.3% of rubric points and GPT-6 Luna earned 91.8%, while costing 98% less, at 8.5 cents per task. Both models were evaluated at maximum reasoning effort. A 99% safety rating on 325 tasks means the reviewers did not rate every task clinically safe.

OpenAI names four partners in the posts:

  • CVS Health: one agent cleared a months-long document-sending backlog over a weekend. Another handles the system steps for tens of thousands of consults each day. CVS is open-sourcing its Agent Kit, built with OpenAI's Agents SDK. The post says CVS production data stays in CVS-controlled infrastructure and is not used to train foundation models.

  • Cigna Healthcare: case managers previously navigated more than 10 systems. OpenAI says the move from concept to deployment took under eight weeks.

  • Abridge: planned inpatient use, with an example of an assistant asked to help resolve discharge barriers.

  • Ambience: in the demonstration, the agent stops and asks the clinician when the record is ambiguous, and waits for approval before submitting.

Trusted clinical search was updated in ChatGPT for Healthcare and ChatGPT for Clinicians. OpenAI says more than 1,300 physician reviews in a blind evaluation rated the update higher on all eight dimensions, and names factual correctness, safety, actionability, and source authoritativeness among them.

The healthcare post lists 21 new plugins, available in ChatGPT for Healthcare, including through ChartGPT in the electronic-health-record workflow: Aiva, Amion, Assort Health, Availity Extend, Elation Health, Ferry, IMO Health MCP Server, Infinitus Relay, K Health, Luma Health, Metriport, Notable Integration Specs, Oracle HDI Overview, PenguinAI Medical Coding, Photon, Redox, Stedi, Subsalt, Trizetto, Turquoise, and UpToDate Expert AI. The page's own examples use TriZetto for an eligibility check and name Aiva, Amion, Metriport, Stedi, and UpToDate Expert AI in sample prompts.

What people are saying

  • "Computer-use agents, improved trusted clinical search, and new plugins are available today in ChatGPT for Healthcare, ChatGPT for Clinicians, and the OpenAI API." — Karan Singhal ↗ (@thekaransinghal), OpenAI Health

  • "Less time on administrative work means more time for care. That's a win for our teams and everyone we serve." — Lagan Khare, Executive Director, Engineering, CVS Health, in OpenAI's post ↗

  • "Moving from concept to deployment in under eight weeks shows what's possible when we bring together AI, our expertise in care management, and the right guardrails for responsible use." — Madhu Nutakki, CIO, Cigna Healthcare, in the same post

  • "OpenAI says the agents can read screens, click, and type to move work between existing apps, with human review on the steps an organization chooses" — artem ↗ (@sunset_evm)

Our take

Prior authorization is the right first target. The work sits in portals that do not expose a complete API: a record in one system, a form in another. Ambience's demonstration prepares the insurer request and stops when the chart is ambiguous. That is a narrower job than handing a model a diagnosis.

Read the percentages as OpenAI's evaluation, and keep a person on the step that leaves the building. The inbox study and HealthAdminBench are OpenAI's report. OpenAI calls HealthAdminBench an independent Stanford benchmark. The 99% figure still leaves tasks the reviewers did not call clinically safe. Approval steps are configurable, and Ambience's demo waits before submit. HIPAA coverage, on OpenAI's page, is the business associate agreement plus the security and retention settings.

The same calendar day is a different case. California sued OpenAI on November 16 over alleged escapes from research-testing sandboxes. That complaint is about lab containment. It is not a finding about ChatGPT for Healthcare.

Who should try it: U.S. clinicians, on the free ChatGPT for Clinicians account; health IT teams that can turn computer use on; developers on the API. Abridge and Ambience, on OpenAI's account, are building on the same capability. This note does not score anyone else's product. Apple's Health AI study is a research app, not this agent. ChatGPT dots connect computers you own. GPT-6 Astra in the science-blog note is a physics post, not a clinical result. Assistant plans are in ChatGPT vs Claude vs Gemini. Other builders are in the no-code agent builder review.

More from the news desk · About Rai Ansar