THE ESSENTIALS IN 30 SECONDS

  • OpenAI announced on November 16 that its computer-use agents can work inside electronic health records and insurer portals by clicking and typing like a human user.
  • The update adds more than 20 healthcare plugins and an upgraded clinical search in ChatGPT for Healthcare and ChatGPT for Clinicians.
  • CVS Health, Cigna Healthcare, Abridge and Ambience describe early deployments; the safety and benchmark figures come from OpenAI's own evaluations.

OpenAI announced on November 16 a set of healthcare updates built around computer-use agents: AI systems that operate existing software visually, by clicking, typing and moving between screens. The company is pitching them at hospitals, insurers and health-tech firms whose electronic health records (EHRs) and payer portals often lack complete APIs. The details below are based on reports by HIT Consultant and Interesting Engineering, which both covered the announcement the same day.

What is new

According to both reports, the agents run on OpenAI's GPT-6 Astra and GPT-6 Luna models and are available through the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians. In a demonstration described by Interesting Engineering, an agent reviewed a back-pain patient's record, drafted an MRI order, a note and a patient message, then prepared a prior-authorisation request in an insurer's portal, with staff reviewing each document before it went further. Organisations can require approvals, limit agent permissions and keep audit logs.

OpenAI also added more than 20 healthcare plugins to ChatGPT for Healthcare, from vendors including Availity, Elation Health, Luma Health, Redox, Stedi and UpToDate, HIT Consultant reports, plus an upgraded clinical search for ChatGPT for Healthcare and ChatGPT for Clinicians. When OpenAI launched ChatGPT for Clinicians in April as a free tool for verified U.S. physicians, nurse practitioners, physician assistants and pharmacists, it mainly answered questions; with this update, it can also act inside clinical software.

Early users

CVS Health says one agent cleared a months-long backlog of documents to send over a single weekend. Cigna Healthcare's case managers use agents to gather member information from more than 10 systems. Abridge is building agents for inpatient discharge tasks, and Ambience showed prior-authorisation submissions that pause to ask clinicians when records are unclear.

How far to trust the numbers

The headline figures are OpenAI's. Physicians rated 99% of 325 reviewed EHR inbox tasks as safe, and outputs were judged very or extremely helpful in 93% of 646 assessments. On HealthAdminBench, a Stanford benchmark covering prior authorisations and claim appeals, Astra scored 94.3% and Luna 91.8%. These are internal or vendor-reported evaluations, not regulatory approvals or independent audits. The two reports also disagree on which model carries the quoted per-task cost, so pricing remains unclear. OpenAI describes its controls as supporting HIPAA-compliant deployments, which is not a guarantee of compliance.