subagentic.ai RSS X YouTube GitHub
AI-generated illustration

News

OpenAI puts computer-use agents into healthcare software

OpenAI opens computer-use agents for healthcare teams, with stronger clinical search and more than 20 plugins in ChatGPT for Healthcare.

openaicomputer-usehealthcarechatgpt

On November 16, 2026, OpenAI published two product posts that work as one launch: computer-use agents that operate inside the clinical and payer software teams already run, and a ChatGPT for Healthcare update with stronger clinical search and more than 20 plugins.

Those agents read screens, click, type, and move between applications. When a clinician recommends an MRI, the facts an insurer needs may already be in the record, but someone still has to retrieve them and type them into another portal. In a demo built on synthetic patient data, an agent reviews the chart and drafts an order, a note, and a patient reply for a clinician before preparing the authorization request for a specialist to review.

Healthcare teams, developers, and individual clinicians can use the capability through the OpenAI API, ChatGPT for Healthcare, and ChatGPT for Clinicians. Existing customers enable it in ChatGPT Work and Codex through a workspace administrator.

OpenAI says it is working with Abridge, Ambience, Cigna Healthcare, and CVS Health on referrals, prior authorization, discharge planning, and outreach between visits. The deployment details that follow are OpenAI’s account of that work.

CVS Health is the clearest production case. OpenAI says one agent cleared a months-long document-sending backlog over a weekend, and another handles system steps for tens of thousands of consults a day, including steps APIs do not fully support. CVS uses validation checks and human oversight, with screenshots and action records. Data stays inside CVS Health’s enterprise-controlled infrastructure, under its AI governance team, and production data is not used to train foundation models. CVS is open-sourcing Agent Kit, built with the OpenAI Agents SDK. Lagan Khare, executive director of engineering at CVS Health, said: “Less time on administrative work means more time for care. That’s a win for our teams and everyone we serve.”

Cigna Healthcare, as part of The Cigna Group’s broader collaboration with OpenAI, is using the models to assemble member context that case managers used to gather from more than 10 clinical and administrative systems, including benefits, coverage, providers, appointments, and prior interactions. OpenAI says computer use can now bring that information together in seconds. CIO Madhu Nutakki said the team moved from concept to deployment in under eight weeks.

The other named partners are earlier. Abridge is developing agents for daily applications that lack APIs. In a planned inpatient use case, a clinician can ask the Abridge Assistant for help with discharge barriers, and agents can work at the same time on medication access and home-health coordination. The clinician is brought in when a decision needs clinical judgment. Ambience’s published example is a demonstration of controls: the agent prepares a prior-authorization request from the chart, stops when the record is ambiguous, and waits for approval before submitting. Health systems set allowed actions, can pause the agent, and get a step-by-step audit log. OpenAI also describes Assort using computer use in Codex to finish a referral close-out Athena’s API does not cover, marking the referral “visit scheduled” so it dropped off the staff follow-up list.

OpenAI also reported measurements of GPT-6 Astra and GPT-6 Luna. On physician-shaped EHR inbox tasks, reviewers rated 99% of 325 Astra tasks safe, judging whether inaccuracies, omissions, or actions could plausibly harm a patient. Physicians rated the work very or extremely helpful in 93% of 646 assessments. OpenAI also evaluated administrative tasks such as prior authorizations, appeals, and medical-equipment orders. On HealthAdminBench, which it calls an independent Stanford benchmark of common administrative workflows, Astra earned 94.3% of rubric points and Luna 91.8%. Luna cost 8.5 cents per task, 98% less than Astra, with both models run at maximum reasoning effort.

The companion post upgrades trusted clinical search in ChatGPT for Healthcare and ChatGPT for Clinicians: how the product finds, evaluates, and cites medical evidence. In a blind evaluation, physicians reviewed answers to real-world questions across specialties, including multi-turn conversations. In more than 1,300 reviews, they rated the updated experience higher on all eight dimensions OpenAI measured: factual correctness, safety, actionability, calibration, source authoritativeness, locality alignment, temporal alignment, and goal satisfaction. Unite.AI’s recap notes that these are physician assessments of responses, not measured patient outcomes.

More than 20 plugins are available in ChatGPT for Healthcare, including through ChartGPT, which brings ChatGPT into the EHR workflow: Aiva, Amion, Assort Health, Availity Extend, Elation Health, Ferry Health, IMO Health, Infinitus Relay, KMED, Luma Health, Metriport, Notable, Oracle HDI, PenguinAI, Photon, Redox, Stedi, Subsalt, Trizetto, Turquoise, and UpToDate Expert AI. OpenAI’s examples chain them in one conversation. Metriport can retrieve an outside discharge summary while UpToDate Expert AI supplies follow-up monitoring. Aiva can open a hospital insulin-pump policy while Amion finds the covering endocrinologist. IMO Health, Stedi, and Turquoise Health can match a knee MRI code, check eligibility, and compare negotiated rates at two hospitals. Workspace administrators must enable the new plugins. Eligible, verified U.S. clinicians get improved search and select plugins in the free ChatGPT for Clinicians account. Unite.AI also places the launch after OpenAI’s September 1, 2026 Epic integration, which brought authorized patient context into ChatGPT.

For protected health information, organizations must use eligible services covered by a Business Associate Agreement with OpenAI, apply required security and retention settings, and extend those protections to connected systems. ChatGPT for Healthcare provides role-based access controls and audit logs. OpenAI says organization data there is not used to train foundation models. Both products offer a BAA.

Read the two November 16 posts before enabling an agent, and pick one handoff: an authorization packet, a referral close-out, or a pre-visit confirmation list. Write down which applications it may open, which steps stay with a person, and what the audit log must show. U.S. clinicians can sign up for ChatGPT for Clinicians. Healthcare customers should ask a workspace administrator to turn on computer use in ChatGPT Work and Codex, and only the plugins that workflow needs.

Sources