CVS and Cigna Put OpenAI's Computer-Use Agents to Work in ChatGPT for Healthcare
The agents click through health records and payer portals that lack usable APIs; OpenAI also added more than 20 plugins but has disclosed no prices or error rates.
·2 min read
Prefer Nakoda News on Google
OpenAI on Monday added agents that operate hospital and insurer software to ChatGPT for Healthcare, its product for clinical organisations, and named CVS Health, Cigna Healthcare, Abridge and Ambience Healthcare as early deployers, AlphaSignal reported.
The agents rely on computer use. They read what is on the screen, click, type and move between applications, so they can work inside electronic health records, payer portals and older clinical systems that lack usable APIs, without custom integration work. Much of the administrative load in US healthcare sits in those systems, in referrals, document transfers and prior authorisations that staff complete by hand across many screens.
CVS Health runs agents in production to send documents, process referrals and support specialist consultations. One agent cleared a document backlog that had built up over months in a single weekend. CVS is also open-sourcing an Agent Kit built on OpenAI's Agents SDK. Cigna Healthcare uses agents to gather context for member outreach, work that previously required case managers to move across more than 10 clinical and administrative systems. Abridge is developing agents to tackle barriers to discharging hospital inpatients, such as access to medication and home health coordination. Ambience Healthcare showed an agent pausing for a clinician's input when a patient record was ambiguous.
OpenAI also added more than 20 plugins from partners including UpToDate Expert AI, Availity and Oracle HDI. They let teams review patient records and hospital protocols, prepare prescriptions for a clinician to review, coordinate patient outreach, check insurance eligibility and analyse costs. A product called ChartGPT brings ChatGPT into the health record workflow. Workspace administrators must switch the plugins on.
OpenAI said it has also changed how ChatGPT finds, evaluates and cites medical evidence. In a blind evaluation using real clinical questions across several specialties, physicians rated the updated system higher on all eight measures tested, including factual correctness, safety and the authority of sources, across more than 1,300 reviews. The results come from controlled conditions and have not been independently validated.
Enterprise customers get the agents, clinical search and plugins subject to workspace settings and administrator controls. Verified US clinicians can use the agents, clinical search and selected plugins through free ChatGPT for Clinicians accounts. Developers can reach the computer-use capability through the API, ChatGPT Work and Codex. Organisations can buy ChatGPT for Healthcare through OpenAI's sales team or through private offers on AWS Marketplace. OpenAI has not disclosed prices.
Administrators can set which applications an agent may open, which actions need human approval and which events are logged. The product includes role-based access controls and audit logs. OpenAI offers a business associate agreement for work involving protected health information and says content shared with the product is not used to train its models. The healthcare organisation is responsible for configuring those controls and for extending protections to connected systems.
OpenAI has not identified the models behind the agents. It has also not published error rates, how often a human has to step in, completion times or how the agents cope when a software interface changes. Screen-based automation can fail when a system is redesigned, when a pop-up appears or when a session expires. It has not been settled who is liable when an agent files a faulty referral.
By
Nakoda News AI & Tech Desk
AI models, chips, products and research, and what they can and cannot do.
