OpenAI Advances Computer-Use Agents for Healthcare Workflows
OpenAI has introduced new computer-use agents designed to assist healthcare teams by automating repetitive, multi-step workflows across existing software systems. These agents function by clicking, typing, and navigating between applications, a capability now available to developers and clinicians through the OpenAI API, ChatGPT for Healthcare, and ChatGPT for Clinicians.
The technology is being deployed to handle administrative and clinical processes, including prior authorization requests, referrals, discharge planning, and patient outreach. To ensure secure deployment, the platform supports HIPAA-compliant workflows, configurable approvals, and audit logs.
Performance evaluations for the new models, GPT-6 Astra and GPT-6 Luna, yielded positive results. In clinical inbox tasks, GPT-6 Astra was rated safe in 99% of 325 reviewed tasks and deemed "very" or "extremely" helpful in 93% of 646 assessments. For administrative tasks, the models were tested on HealthAdminBench, an independent benchmark from Stanford researchers. GPT-6 Astra achieved 94.3% of available rubric points, while GPT-6 Luna earned 91.8% at a cost of 8.5 cents per task, representing a 98% cost reduction.
Several healthcare organizations are currently piloting or deploying these agents. CVS Health is utilizing the technology to automate document processing and referral management, notably clearing a months-long document-sending backlog over a single weekend. CVS Health is also open-sourcing its Agent Kit, developed with the OpenAI Agents SDK, to provide a repeatable framework for other organizations. The company maintains that all data remains within its enterprise-controlled infrastructure and is not used to train foundation models.
Abridge is developing agents to assist clinicians with care plans, including managing discharge barriers and coordinating medication access. Meanwhile, Cigna Healthcare is integrating OpenAIβs frontier models to consolidate information across more than 10 separate clinical and administrative systems, aiming to provide care teams with a more comprehensive view of member needs and coverage.
For healthcare administrators and IT leaders, the integration of these agents signals a shift toward automating high-volume, fragmented tasks that currently require manual data entry. While the performance metrics on safety and efficiency appear promising, successful adoption will likely depend on maintaining strict human oversight and rigorous validation protocols within enterprise-controlled environments.
Why it matters
The deployment of autonomous agents capable of navigating legacy healthcare software could significantly reduce administrative burdens and operational costs for health systems.
FAQ
What specific tasks can these computer-use agents perform?
They can carry out multi-step workflows such as prior authorization requests, referrals, discharge planning, patient outreach, and document processing by clicking, typing, and navigating between applications.
How does CVS Health ensure data security with these agents?
CVS Health keeps data within its enterprise-controlled infrastructure, governed by its AI Governance team, and does not use production data to train foundation models.
What is the difference between GPT-6 Astra and GPT-6 Luna in terms of performance?
On the HealthAdminBench, GPT-6 Astra earned 94.3% of rubric points, while GPT-6 Luna earned 91.8% at a 98% lower cost per task.
Source: OpenAI News. See the original source for full details.
Scan with your phone
Scan with your phone