OpenAI Teaches ChatGPT for Healthcare to Click Through Hospital Software and Adds 20+ Plugins
To schedule an MRI referral, a clinic employee has to manually move between the patient chart, the schedule, and the insurance portal. OpenAI has taught ChatGPT for Healthcare agents to read the screen and click through these programs themselves, and added 20+ plugins. In a blind review of more than 1,300 cases, physicians rated the updated clinical search higher on all eight criteria. Routine work can be automated without changing software. Pilots: Abridge, Ambience, Cigna Healthcare, and CVS Health are handling referrals and documents. Administrators choose which applications the agent can use and which steps require physician review. ChatGPT for Clinicians is free for verified physicians in the US.
AI-processed from OpenAI Blog; edited by Hamidun News
Imagine a nurse or coordinator who needs to prepare a patient for an MRI. The data sits in the electronic health record (EHR), the hospital's rules are in a separate database, the schedule is in a third system, and the insurance check has to be done on the payer's website. Every switch between windows takes time, and all of it is done by hand. OpenAI has updated ChatGPT for Healthcare to eliminate this running around: agents now work directly in existing programs, clinical search has become more accurate (physicians reviewed it in more than 1,300 cases), and the catalog has been expanded with more than 20 plugins for medical services.
Agents That Work With the Screen
The main novelty is computer-use agents for healthcare. These are AI assistants that perform multi-step work in familiar applications the way a person does: they read the screen, click, type text, and switch between windows. Where a system has no convenient API, this is sometimes the only way to automate anything.
In OpenAI's demo, a physician asks the ChatGPT app by voice for help with a repeat MRI request. The agent reviews the patient chart, drafts the referral and a reply to the patient. The decision stays with the physician: the drafts go to them for review.
OpenAI talks about more capable, safer, and cheaper models for clinical and administrative tasks. For secure deployment, it has added access controls and audit logs. The organization itself chooses which applications the agent can access and decides which steps a person must verify. Abridge, Ambience, Cigna Healthcare, and CVS Health are already deploying the agents: they process referrals, move documents between systems, and gather information for clinical teams.
Search With Verifiable Sources
The second update concerns clinical search: how ChatGPT finds, evaluates, and cites medical data. It has been updated in ChatGPT for Healthcare and ChatGPT for Clinicians. The goal is for a physician to get an answer together with a practical next step and links that can be opened and checked.
OpenAI shared the results of a blind review. Physicians assessed answers to real clinical questions from different specialties, including multi-turn conversations. In more than 1,300 reviews, the updated version scored higher on all eight parameters, including factual accuracy, safety, practical applicability, and source authority. A caveat: this is the company's own evaluation, the material contains no independent verification, and OpenAI does not disclose the specific scores.
More Than 20 Plugins
The catalog has added more than 20 plugins from companies in clinical care, care coordination, and operations: Luma Health, Metriport, Stedi, Turquoise Health, UpToDate Expert AI, Oracle HDI, Redox, Notable, Amion, IMO Health, and others. They are available in ChatGPT for Healthcare, including through ChartGPT, which embeds ChatGPT into the EHR workflow.
Examples from the announcement show how the plugins work within a single request:
- A coordinator finds tomorrow's visits that have not been confirmed, checks patients' insurance, and gets a list of calls to make;
- A primary care physician finds the discharge summary after a hospitalization for heart failure, sees which medications changed, and gets monitoring recommendations from UpToDate;
- A navigator compares two hospitals for a knee MRI: selects the procedure code, checks insurance coverage, and looks at negotiated rates.
Privacy and Availability
To work with HIPAA-protected data, organizations must use services covered by a BAA with OpenAI, enable the necessary security and retention settings, and extend protection to the connected systems. Content from ChatGPT for Healthcare is not used to train models.
Computer-use agents in ChatGPT Work and Codex and the improved clinical search are available now, subject to workspace settings. An administrator enables plugins. ChatGPT for Clinicians is free for verified physicians in the US. Computer use is available to developers in the OpenAI API, and for enterprise scenarios the company asks customers to contact its sales team.
What It Means in Practice
The healthtech market is shifting from a chatbot with citations to an agent that works inside the hospital stack. The winners will be those with clear access boundaries, audit trails, and mandatory human review of critical steps. An agent that clicks through interfaces is riskier than a regular API integration: it can make a mistake in a window that changed after an update. That is why pilots should sensibly start with administrative tasks such as referrals and insurance checks, while clinical decisions stay with the physician.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What are the key details on Agents That Work With the Screen?
The main novelty is computer-use agents for healthcare. These are AI assistants that perform multi-step work in familiar applications the way a person does: they read the screen, click, type text, and switch between windows. Where a system has no convenient API, this is sometimes the only way to automate anything. In OpenAI's demo, a physician asks the ChatGPT app by voice for help with a repeat MRI request. The agent reviews the patient chart, drafts the referral and a reply to the patient.
What are the key details on Search With Verifiable Sources?
The second update concerns clinical search: how ChatGPT finds, evaluates, and cites medical data. It has been updated in ChatGPT for Healthcare and ChatGPT for Clinicians. The goal is for a physician to get an answer together with a practical next step and links that can be opened and checked. OpenAI shared the results of a blind review. Physicians assessed answers to real clinical questions from different specialties, including multi-turn conversations.
What are the key details on More Than 20 Plugins?
The catalog has added more than 20 plugins from companies in clinical care, care coordination, and operations: Luma Health, Metriport, Stedi, Turquoise Health, UpToDate Expert AI, Oracle HDI, Redox, Notable, Amion, IMO Health, and others. They are available in ChatGPT for Healthcare, including through ChartGPT, which embeds ChatGPT into the EHR workflow.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.