OpenAI Brings Computer-Use Agents to Healthcare
OpenAI adds agents that navigate healthcare software, more than 20 plugins and stronger clinical search, with human review and access controls. Photo: Tessy Agbonome / Pexels
Enterprise Tech

OpenAI Brings Computer-Use Agents to Healthcare

OpenAI adds agents that navigate healthcare software, more than 20 plugins and stronger clinical search, with human review and access controls.

By Daniel Mercer • 3 mins read Edited by Samantha Reed Published: Updated:

Key Notes

  • OpenAI is extending healthcare AI from answering questions to completing work across clinical software under configured review and access controls.
  • The update adds more than 20 healthcare plugins and is available through eligible ChatGPT products and the API.
  • Company benchmark results measure selected tasks and reviewer judgments rather than demonstrated improvements in patient outcomes.

OpenAI is bringing computer-use agents to healthcare teams, expanding ChatGPT from a source of answers into software that can help carry work across medical records and insurers’ portals. The company announced the update on November 16 alongside improved clinical search and more than 20 healthcare plugins.

The commercial opportunity is the work between systems: finding the right record, preparing documentation and entering information in another application. Completing that chain reliably is a different challenge from writing a convincing summary. In healthcare, a missed step can delay care even when the generated text is correct.

From a Patient Message to an Authorization Request

In a demonstration using synthetic patient data, an agent reviews records relevant to an MRI request, drafts an order and patient response, then prepares an insurer’s authorization form. The drafts are presented for professional review. The demonstration illustrates the intended workflow; it is not evidence of an autonomous treatment decision involving a real patient.

OpenAI’s accompanying research update reports that physicians rated 99% of 325 reviewed GPT-6 Astra tasks clinically safe. Separately, 93% of 646 helpfulness assessments placed its work in the top two categories. Those are judgments about selected tasks and outputs, rather than a measured reduction in patient harm.

On HealthAdminBench, OpenAI reports Astra earned 94.3% of available rubric points and GPT-6 Luna earned 91.8%, with Luna costing 8.5 cents per task at maximum reasoning effort. A rubric score is not the percentage of complete workflows performed perfectly.

The benchmark’s authors describe 135 tasks covering prior authorizations, medical-equipment orders and appeals. Agents interact with simulated administrative portals, and evaluation combines checks on application state with other scoring criteria. That makes the benchmark useful for comparing workflow capability while leaving production reliability to be established in deployments.

Plugins Connect More of the Existing Workflow

The new plugins cover functions such as retrieving records, checking insurance eligibility, coordinating outreach and examining costs. OpenAI says they are available in ChatGPT for Healthcare, subject to administrators enabling them. Its clinical-search update also improves how the system retrieves and cites medical evidence.

The broader industry is pursuing the same transition from documentation to action. In a separate announcement Monday, healthcare AI developer Abridge described an assistant that coordinates work across fragmented systems and said its experience is live in early environments with a few selected partners. That is a narrower deployment claim than general availability across its customer base.

Access Controls Remain Part of the Product

OpenAI says computer use is available through its API, ChatGPT for Healthcare and ChatGPT for Clinicians. The clinicians product is free for eligible, verified US clinicians. Organizations handling protected health information need eligible services covered by a Business Associate Agreement and the required security and retention settings; OpenAI says organizational data is not used to train its foundation models.

Access permissions, audit trails and approval steps determine how much authority an agent receives. For buyers, the practical test is whether the system completes the intended task, leaves a usable record and brings a person back in when judgment is required. AIstify’s recent coverage of competing healthcare agents shows how central that workflow question has become to enterprise AI.

Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.

AI & Machine Learning, Enterprise Tech, News