Kivorent News

Wednesday, November 18, 2026

Health techNews

OpenAI adds computer-use agents to ChatGPT for Healthcare

OpenAI added AI agents that click through electronic health records and payer portals, plus more than 20 health tech plug-ins, to ChatGPT for Healthcare at HLTH on November 16. CVS Health and Cigna are early users.

Docket, covering health regulation and AI in care
, 3 min read

99%Agent actions rated clinically safe in OpenAI's own test

What to know

  • OpenAI's computer-use agents, powered by GPT-6 Astra and GPT-6 Luna, operate electronic health records (EHRs) and insurer portals on screen where no API exists, according to HIT Consultant.
  • ChatGPT for Healthcare gained more than 20 plug-ins, including Availity Extend, Redox, Stedi, TriZetto, Photon and UpToDate Expert AI.
  • The safety and accuracy figures come from OpenAI's own evaluations and one Stanford benchmark; none is a clinical trial or an FDA review.

OpenAI has added computer-use agents to ChatGPT for Healthcare, its product for hospitals and health companies. The agents work inside existing software the way a person does, reading the screen, clicking and typing, so they can operate electronic health records (EHRs) and insurer portals that have no usable API. OpenAI announced the update on November 16 at the HLTH conference, according to HIT Consultant, which reported the details. Healthcare IT Today's HLTH roundup confirmed the agents and the new plug-ins. OpenAI's own announcement page could not be opened for this story.

What OpenAI added

The agents run on OpenAI's GPT-6 Astra and GPT-6 Luna models. HIT Consultant reports they are available now through the OpenAI API, ChatGPT for Healthcare, ChatGPT Work, Codex and ChatGPT for Clinicians, which is free for eligible US clinicians.

In OpenAI's example, an agent works through an EHR inbox: it reviews a child's fracture notes and weight log, then drafts ibuprofen dosing instructions for the caregiver, cast-care guidance and a chart note. Nothing is sent or signed until a clinician approves it. OpenAI describes human checkpoints, action-by-action audit logs, agents that stop when a clinical question is ambiguous, and no training of its models on customers' data.

ChatGPT for Healthcare also gained more than 20 plug-ins from health tech companies. HIT Consultant lists 21, including Availity Extend, Amion, Elation Health, IMO Health, Infinitus, K Health, Luma Health, Metriport, Oracle HDI, Photon, Redox, Stedi, TriZetto and Wolters Kluwer's UpToDate Expert AI. OpenAI also upgraded clinical search in ChatGPT for Healthcare and ChatGPT for Clinicians.

Who is using it

CVS Health's engineering teams use the agents for sending documents, processing referrals and arranging specialist consults, according to HIT Consultant. One agent cleared a months-long backlog of documents over a weekend, and CVS plans to open-source the agent kit it built on OpenAI's Agents SDK. Cigna Healthcare took a case-manager outreach tool from concept to production in under eight weeks. Abridge is building inpatient agents for discharge delays, and Ambience showed an agent submitting prior authorization requests on insurer portals.

How the numbers were measured

The headline figures are OpenAI's. On physician-designed EHR inbox tasks that OpenAI ran, GPT-6 Astra's actions were rated clinically safe 99 percent of the time across 325 reviewed tasks, and physicians rated the output very or extremely helpful in 93 percent of 646 assessments. On HealthAdminBench, a benchmark from Stanford researchers covering prior authorizations, claim appeals and equipment orders, Astra scored 94.3 percent and Luna 91.8 percent, at 8.5 cents per task for Luna. In a blind review of more than 1,300 clinical conversations, OpenAI says clinical search's factual correctness rose from 93.4 to 97.6 percent.

These are vendor tests, not randomized trials or regulatory reviews. HIT Consultant does not report whether OpenAI offers a HIPAA business associate agreement for the new features or what they cost. For how such evidence differs from a trial, see our explainer on randomized controlled trials.

Why it matters

Much of the paperwork in US health care, from prior authorization to referrals, runs through portals and EHR screens that were never built for automation. Agents that can work those screens without new integrations could reach that work faster than API-based tools, which puts OpenAI in direct competition with the scribe and automation startups that sell to hospitals. It arrives as federal regulators push electronic prior authorization (ONC and CMS testing) and CMS lays out its own AI roadmap. Tools that draft clinical instructions also raise the question of when software becomes a regulated device, covered in our AI scribe explainer. OpenAI's consumer health push began with ChatGPT Health in January, and Apple is now testing its own health chat app.

This article is general information about how technology works. It is not medical advice. Talk to a clinician about your own health.

Sources

OpenAIAI agentsDigital health and AI in careAI models

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.