OpenAI said Monday that ChatGPT for Healthcare can now use computer-use agents to carry out multi-step work inside the software clinics and insurers already run, by reading screens, clicking, typing and moving between applications.

The Nov. 16 product post said the agents are meant to help teams move information between disconnected systems. In a demonstration that OpenAI said used synthetic patient data, an agent reviews a chart and drafts an MRI order and a patient reply for a clinician to review, then prepares a prior-authorization request in an insurer’s portal for an authorization specialist to review. The same announcement added improved “trusted clinical search” and “more than 20 new healthcare plugins,” which OpenAI said are available in ChatGPT for Healthcare. It listed 21 plugins by name.

OpenAI named Abridge, Ambience, Cigna Healthcare and CVS Health as organizations it is working with on computer use “across authorization requests, referrals, discharge planning, and outreach between visits.” A companion post described those efforts at different stages. None of the four issued a newsroom release on the Nov. 16 computer-use launch.

For CVS Health, OpenAI said the company “is using” the capabilities to send documents, process referrals and support specialist consults, including in “one production workflow” that “cleared a months-long document-sending backlog over a weekend,” and another that “handles the system steps for tens of thousands of consults each day,” with validation checks, human oversight, screenshots and action records. OpenAI said CVS is open-sourcing an Agent Kit on GitHub.

OpenAI said Abridge “is developing” computer-use agents to help clinicians carry out care plans, and described a “planned inpatient use case” for discharge barriers such as medication access and home-health coordination. Abridge’s press page listed a Nov. 16 Wiley partnership release, not one on these agents. For Cigna Healthcare, OpenAI said computer use “can now help” case managers gather context from “more than 10 separate clinical and administrative systems” in seconds. It quoted chief information officer Madhu Nutakki as saying the work moved “from concept to deployment in under eight weeks.” A Sept. 23, 2026, Cigna Group release on a broader OpenAI collaboration did not describe computer-use agents; it said “clinicians remain responsible for patient care and clinical decisions.” OpenAI presented Ambience’s work as a demonstration: an agent prepares a prior-authorization request, “stops and asks the clinician, then waits for approval before submitting.” Health systems, it said, set permitted actions, can pause the agent and get a step-by-step audit log.

OpenAI said organizations can “set which applications agents can access, and establish clear controls for which steps require human review,” and that ChatGPT for Healthcare provides role-based access controls and audit logs. The agents are available in ChatGPT Work and Codex, “subject to workspace settings.” For HIPAA-compliant workflows involving protected health information, OpenAI said organizations “must use eligible services covered by their BAA with OpenAI, apply required security and retention settings, and extend protections to connected systems.” Both ChatGPT for Healthcare and the free ChatGPT for Clinicians product offer a Business Associate Agreement. OpenAI said content shared with ChatGPT for Healthcare is not used to train models. A current OpenAI help article lists “Use Windows computer-use” among Codex functions covered by a BAA when used from an eligible workspace, and lists “Allow browser use for Work in the cloud” among functions that are not covered. ChatGPT for Healthcare’s help page says, “Final decisions always remain with the clinician.”

OpenAI reported two separate evaluations, both company claims. To measure the clinical-search updates, it said that “in a blind evaluation, physicians reviewed responses to real-world clinical questions across multiple specialties, including multi-turn conversations,” and that “in over 1300 reviews, physicians rated the updated experience more highly across all eight dimensions evaluated, including factual correctness, safety, actionability, and source authoritativeness.” That evaluation is of clinical search, not the agents. Separately, OpenAI said it “evaluated GPT-6 Astra on tasks clinicians handle in their EHR inboxes,” with tasks “shaped by physician input,” and assessed safety by “whether inaccuracies, omissions, or actions in the model’s work could plausibly lead to patient harm.” It said “99% of 325 reviewed tasks were rated safe,” and that physicians selected “very helpful” or “extremely helpful” in “93% of 646 assessments.” On HealthAdminBench, which it called “an independent benchmark from Stanford researchers,” OpenAI reported GPT-6 Astra “earned 94.3% of available rubric points” and GPT-6 Luna “earned 91.8%, despite costing 98% less at 8.5 cents per task.”

In an interview with Fierce Healthcare published Monday, Karan Singhal, OpenAI’s head of Health AI, said researchers did not identify any cases where the model created a specific, plausible risk of patient harm. “In some cases, the model might overreach and might take an action without getting the proper sign-off or authorization,” he said. OpenAI provides the technology and safeguards, he said, while partners handle “the last mile” of implementation.

OpenAI’s product posts do not assign legal liability if an agent submits a wrong referral or order. Its computer-use developer documentation tells integrators to “ask the user immediately before” actions that include “taking medical-care actions,” and to “keep a human in the loop for purchases, authenticated flows, destructive actions, or anything hard to reverse.” The FDA’s January 2026 Clinical Decision Support Software guidance explains which functions the agency considers excluded from the device definition. OpenAI did not address the agents’ device status. As of Tuesday morning, no statements responding to the launch were found from the FDA or HHS.