OpenAI Adds Computer-Use Agents and 20+ Plugins to ChatGPT for Healthcare

OpenAI’s latest ChatGPT for Healthcare update adds computer-use agents, more than 20 plugins, and improved clinical search. Here’s what healthcare organizations can automate and what IT teams need to verify.

Nov 19, 2026
OpenAI Adds Computer-Use Agents and 20+ Plugins to ChatGPT for Healthcare

OpenAI's latest ChatGPT for Healthcare update introduces computer-use agents and more than 20 plugins to help clinical teams work across healthcare software, with human oversight and access controls. Image generated by ChatGPT.

We may earn from vendors via affiliate links or sponsorships. This might affect product placement on our site, but not the content of our reviews. See our Terms of Use for details.

An AI agent that can move between hospital systems, retrieve patient information, and prepare an MRI request could cut some of the repetitive work behind patient care. OpenAI is bringing that capability to ChatGPT for Healthcare, alongside more than 20 new plugins and improved clinical search.

Announced Nov. 16, the update introduces computer-use agents that navigate existing software by reading screens, clicking, and typing. The agents are designed to help clinical and administrative teams work across electronic health records (EHRs), scheduling applications, and insurance portals.

The expansion also raises questions about which systems these agents can access, what actions they can perform, and when a clinician must step in. OpenAI provides administrative controls and audit logs, but healthcare organizations still need to configure permissions and review connected services.

What OpenAI’s new healthcare agents can do

Advertisement

In OpenAI’s announcement, computer-use agents are designed to complete multistep workflows across applications healthcare workers already use.

In one demonstration using synthetic patient data, a clinician asks an agent to follow up on an MRI request. The agent reviews the patient’s chart, drafts an imaging order and patient response, and prepares information for an insurance authorization request.

The proposed documents and actions are then reviewed by healthcare professionals.

OpenAI is also working with organizations including Abridge, Ambience, Cigna Healthcare, and CVS Health on workflows involving referrals, authorization requests, discharge planning, and patient outreach.

Those partnerships and demonstrations don’t mean every healthcare organization can immediately automate the same tasks.

Advertisement

More than 20 new healthcare plugins

OpenAI’s expanded plugin ecosystem connects ChatGPT with applications used for clinical care, care coordination, and business operations.

The additions include UpToDate Expert AI, Redox, Metriport, K Health, Availity Extend, Stedi, TriZetto, and PenguinAI Medical Coding.

Their functions differ by provider.

  • Metriport: Retrieves outside medical records and prepares source-linked summaries for patient visits.
  • UpToDate Expert AI: Provides clinical guidance for healthcare professionals reviewing treatment questions.
  • Stedi and TriZetto: Help check insurance eligibility and coverage.
  • Luma Health: Supports appointment management, waitlists, and follow-up activities.
  • Turquoise Health: Compares negotiated healthcare prices across hospitals and payers.

Some plugins also support actions beyond retrieving information.

Photon, for example, allows verified clinicians to draft, review, and send prescriptions through a connected account. Each prescription requires prescriber authorization, and controlled substances aren’t supported.

Advertisement

The new plugins are available through ChatGPT for Healthcare, including ChartGPT, which brings ChatGPT into EHR workflows.

TechRepublic previously covered ChatGPT’s Epic integration, which gives authorized users read-only access to supported patient records.

Clinical search gets an upgrade

OpenAI has also improved trusted clinical search, which retrieves medical evidence and provides citations that clinicians can inspect.

In a blind evaluation involving more than 1,300 physician reviews, the company says the updated experience received higher ratings across eight dimensions, including factual correctness, safety, actionability, and source authoritativeness.

The findings are company-reported, and the announcement doesn’t establish that they were independently audited.

OpenAI separately evaluated computer-use agents on clinical tasks. Its research found that 99% of 325 reviewed tasks were rated clinically safe, although the evaluations don’t establish equivalent performance in live patient-care settings.

Advertisement

Who can access the new features?

Availability differs across OpenAI’s offerings.

ChatGPT for Healthcare: Existing organizational customers can access computer-use agents through ChatGPT Work and Codex, along with improved clinical search. Administrators must enable the appropriate capabilities and plugins.

ChatGPT for Clinicians: Eligible, verified US clinicians can use computer-use capabilities and updated clinical search through the free clinician offering. Selected plugins are also available.

Developers: Computer-use capabilities are available through the OpenAI API, ChatGPT Work, and Codex for teams building clinical and administrative workflows.

Individual plugin access depends on workspace settings, connected accounts, provider requirements, and user permissions.

Advertisement

What healthcare IT teams should check before deployment

Computer-use agents can interact with healthcare applications in ways that go beyond summarizing documents.

An agent navigating an EHR or insurance portal may encounter protected health information (PHI), enter data, or prepare actions across several systems. The permissions that govern those activities need to be defined before deployment.

Permissions and human approval

Healthcare IT teams should review five areas before expanding agent access:

  • Application access: Which EHRs, scheduling systems, and insurance portals can agents reach?
  • Read and write permissions: Can an integration retrieve information, modify records, or submit transactions?
  • Human review: Which actions require approval from a clinician or administrator?
  • Accountability: How are agent activities, approvals, and errors recorded?
  • Vendor access: What agreements and permissions govern connected third-party services?

TechRepublic’s coverage of AI agent governance examines how organizations assign responsibility for automated actions and monitor sensitive data access.

Advertisement

The National Institute of Standards and Technology has also warned about the risks of credential sharing, long-lived access tokens, and overly broad permissions.

In its August guidance, NIST recommends building on established identity and authorization standards while developing stronger protections for autonomous systems.

HIPAA and patient data protection

OpenAI offers Business Associate Agreements (BAAs) for eligible ChatGPT for Healthcare and ChatGPT for Clinicians deployments to support HIPAA-compliant workflows.

However, a BAA doesn’t automatically cover every feature or connected application.

Healthcare organizations must use eligible functionality covered by their agreements, configure required security and retention settings, and confirm that appropriate protections extend to connected services.

OpenAI’s documentation also notes that improved memory isn’t covered by the Healthcare BAA. Organizations shouldn’t enter PHI when using that feature.

Individual clinicians must have an applicable BAA in place and be authorized to use it before entering PHI into ChatGPT for Clinicians.

Advertisement

OpenAI says content shared with ChatGPT for Healthcare isn’t used to train its models.

Those commitments don’t replace an organization’s responsibility to review how patient information moves between applications.

TechRepublic previously examined enterprise AI access controls and the security risks associated with agents operating under existing user permissions.

Where organizations should start

A limited pilot can help healthcare teams evaluate computer-use agents before granting broader access to production systems.

Preparing referral summaries, retrieving authorized documents, and drafting patient communications offer opportunities to measure accuracy, permissions, and the amount of human intervention required.

Teams can track preparation time, corrections, unsuccessful actions, and whether employees need to repeat work the agent was supposed to complete.

Clinical decisions, prescription authorization, and changes to sensitive records should remain subject to appropriate professional review.

Advertisement

OpenAI’s new capabilities could reduce administrative work across disconnected healthcare systems. The next test for hospitals is whether those workflows remain reliable when agents encounter real-world software restrictions, incomplete records, and the approval requirements of patient care.

Also read: OpenAI is bringing free ChatGPT training to caregivers in five US cities, covering everyday tasks, organizing information, and responsible AI use.