
The Library Media — Science, Technology & Health · AI News & Intelligence Desk
By Maya Wilson, World Desk Lead, The Library Media
OpenAI has made computer-use agents available for healthcare work, bringing software agents that can read screens, click and type into the clinical and payer systems already used by care teams. The company says the agents can carry out multi-step administrative and clinical-support workflows across disconnected applications, including authorization requests, referrals, discharge planning and outreach between visits.
The announcement, detailed in two OpenAI product updates published November 16, extends computer-use capabilities across the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians. The agents are powered by GPT-6 Astra, with GPT-6 Luna positioned as a lower-cost model for high-volume administrative tasks. Existing customers can enable the capability in ChatGPT Work and Codex through a workspace administrator.
OpenAI says it is working with Abridge, Ambience, Cigna Healthcare and CVS Health. CVS provides the company’s clearest production examples. According to OpenAI, one agent cleared a document-sending backlog that had accumulated over several months in a single weekend. Another handles system steps for tens of thousands of specialist consultations each day, including processes that are not fully supported by application programming interfaces.
CVS uses validation checks and human oversight for the workflows, with screenshots and records of agent actions. The company is also open-sourcing its agent kit, which was built using the OpenAI Agents SDK.
At Cigna Healthcare, computer use assembles member context for case managers who previously had to collect information across more than 10 separate clinical and administrative systems. The information includes benefits, coverage, providers, appointments and prior interactions. OpenAI says Cigna moved from concept to production in less than eight weeks.
Abridge is developing computer-use agents to help clinicians carry out care plans. One planned inpatient application focuses on discharge barriers.
OpenAI also published evaluation results for its models. In the company’s study of multi-step electronic-health-record inbox workflows, physicians rated 99% of 325 GPT-6 Astra tasks as clinically safe. In a separate set of 646 assessments, physicians rated the work very helpful or extremely helpful 93% of the time.
On the HealthAdminBench administrative benchmark, Astra scored 94.3%, while Luna scored 91.8%. OpenAI reported a cost of about 8.5 cents per task for Luna.
Those figures describe evaluations of model performance and physician assessments of the agents’ work. They are not measurements of clinical outcomes in patients, nor do they establish that the agents improve patient health. The results reflect how the systems performed on the evaluated tasks and how physicians judged that work.
OpenAI says organizations control which applications agents can access and can require human review at designated steps. Agent actions are recorded in audit logs. For HIPAA-covered use, organizations must have a Business Associate Agreement in place and use the required security settings. OpenAI says healthcare data is not used to train its models.
The broader ChatGPT for Healthcare update also adds stronger clinical search and more than 20 healthcare plugins. ChatGPT for Clinicians is free for eligible, verified clinicians in the United States. OpenAI did not announce pricing for the healthcare agent capability.
Maya Wilson
World Desk Lead, The Library Media
From The Library — the estate of our family.
Follow Us