Innovation & Tech

OpenAI's Health Agents Are Already Clearing Backlogs at CVS

OpenAI's healthcare computer-use agents are live: 99% of reviewed GPT-6 Astra EHR tasks rated clinically safe, and CVS Health agents already clear tens of thousands of consults daily.

By Nathan Brooks

4 min read

OpenAI Brings Computer-Use Agents to Healthcare Workflows
OpenAI Brings Computer-Use Agents to Healthcare WorkflowsAI-generated

What's News

  • OpenAI announced healthcare computer-use agents on November 16, 2026; 99% of 325 reviewed GPT-6 Astra EHR inbox tasks were rated clinically safe.
  • GPT-6 Luna ran at 8.5 cents per task on HealthAdminBench — 98% cheaper than GPT-6 Astra, which scored 94.3% of rubric points.
  • A CVS Health production agent cleared a months-long document backlog over a weekend; another handles tens of thousands of specialist consults daily.
  • Cigna moved its case-manager information assembly from concept to deployment in under eight weeks.
  • CVS Health open-sourced its CUA Agent Kit under Apache-2.0, converting agent runs into model-free Playwright or pywinauto code.

By Nathan Brooks

OpenAI's computer-use agents cleared 99% of their safety review. The company announced healthcare agents on November 16, 2026, and reported that reviewers rated 325 reviewed GPT-6 Astra tasks in electronic health record inboxes as clinically safe — meaning inaccuracies, omissions or actions in the model's work couldn't plausibly lead to patient harm.

The agents are available through ChatGPT for Healthcare, ChatGPT for Clinicians and the OpenAI API. They don't need new integrations. They operate existing software interfaces to complete multistep work: authorization requests, referrals and discharge planning.

In a synthetic-patient demo, a message about lower back pain triggered review of visit notes, physical-therapy records and prior imaging. The agent drafted an MRI order, encounter note and patient reply for clinician review, then prepared an insurer-portal authorization request for a specialist to sign off.

What do the benchmarks actually show?

On HealthAdminBench, an independent benchmark from Stanford researchers, GPT-6 Astra earned 94.3% of available rubric points and GPT-6 Luna earned 91.8%. The benchmark spans 135 tasks and 1,698 subtasks covering prior authorization, durable medical equipment orders and appeals.

The cost gap is the number worth pinning to your wall. Luna ran at 8.5 cents per task — 98% cheaper than Astra — with both models at maximum reasoning effort. For high-volume administrative work, that's the difference between a pilot and a line item.

Physicians also weighed in. OpenAI reports doctors selected the two highest helpfulness categories — very helpful or extremely helpful — in 93% of 646 assessments on a five-point scale.

Production work: CVS Health and Cigna

This isn't a demo stage anymore. According to OpenAI, one CVS Health production agent cleared a months-long document-sending backlog over a weekend. Another handles system steps for tens of thousands of specialist consults every day.

The agents navigate existing applications to complete steps that APIs don't fully support. CVS Health validates with checks, human oversight, screenshots and action records. Data stays inside its enterprise-controlled infrastructure, and production data isn't used to train foundation models.

CVS Health also open-sourced its CUA Agent Kit under an Apache-2.0 license. The kit uses an OpenAI computer-use agent to discover browser or Windows workflows, then converts successful runs into inspectable Playwright or pywinauto code that runs without model calls. Translation: stable workflows stop burning per-task model spend, while Codex-assisted diagnosis and updates stay subject to engineer review, testing and verification.

Cigna Healthcare's case managers previously navigated more than 10 separate clinical and administrative systems to gather benefits, coverage, provider, appointment and prior-interaction context. Its deployment assembles that information for member outreach and moved from concept to production in under eight weeks. The work sits inside The Cigna Group's broader OpenAI collaboration.

Human review, escalation and access controls

Abridge is building a cross-system assistant that coordinates discharge work across medication access and home health, with concurrent agents and escalation for clinical judgment. Its announcement says the experience is live in early environments with a few select partners, and that oversight must show what information was used, whether each step completed, when the assistant escalated and whether a payer received a submission.

Ambience's demonstration applies the same discipline to prior authorization. The agent prepares a request from chart evidence, stops to ask a clinician when the record is ambiguous, and waits for approval before submitting. Health systems can restrict permitted actions, pause the agent and inspect a step-by-step audit log.

What does this change for operators?

Two things, and one warning. First, ChatGPT for Healthcare ships with role-based access controls and audit logs, and organizations can add approval steps and limit agents to the applications, records and actions each task requires. For HIPAA-compliant workflows involving protected health information, organizations must use eligible services covered by an OpenAI Business Associate Agreement plus the required security and retention settings.

Second, the warning: CVS Health's own repository cautions that run artifacts are not sanitized exports. Traces can contain credentials, cookies, session tokens and network data, and screenshots capture everything visible on screen. Credential replacement and redaction markers are best-effort protections, not comprehensive redaction.

The split between CVS Health's production administrative agents and Abridge's early care-coordination environments defines where this market actually is. Computer use is already handling high-volume interface work today. Broader clinical coordination will move only as fast as audit trails, access limits and escalation points get built — and that's where the next competitive gap opens.

Source: unite.ai

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.

2,231 articles

Related articles

« Previous articleNext article »