Your Chief AI Officer · Tuesday, November 17, 2026 · Act today

OpenAI builds computer-use agents for referrals and prior authorization

Announced3 sourcesFor employees and patients

OpenAI is building computer-use agents with CVS Health, Cigna Healthcare, Abridge and Ambience Healthcare to handle referrals, prior authorizations, discharge planning and other work across existing healthcare software, Becker's reported. The agents, announced Nov. 16, read screens, click, type and move between applications such as EHRs and payer portals, and health systems can access them through ChatGPT. HIT Consultant reports the release also adds 20+ integrated health tech plugins.

OpenAI is building computer-use agents with CVS Health, Cigna Healthcare, Abridge and Ambience Healthcare to handle referrals, prior authorizations, discharge planning and other work across existing healthcare software, Becker's reported. The agents, announced Nov. 16, read screens, click, type and move between applications such as EHRs and payer portals. Health systems can reach them through ChatGPT for Healthcare, the free ChatGPT for Clinicians and the OpenAI API. HIT Consultant reports the release also adds more than 20 healthcare plugins, including Availity Extend, Oracle HDI Overview, Redox and UpToDate Expert AI.

The early deployments point at the work that consumes staff time. At CVS Health, one agent cleared a monthslong document backlog over a weekend and another handles the system steps for tens of thousands of consults each day, according to Becker's. Cigna case managers who previously navigated more than 10 systems before member outreach now have that information gathered in seconds, and Cigna CIO Madhu Nutakki said the team moved from concept to deployment in under eight weeks. CVS is open-sourcing its Agent Kit, built on OpenAI's Agents SDK.

For finance and revenue cycle leaders, the economics are the headline. On HealthAdminBench, an administrative benchmark from Stanford researchers covering prior authorizations, claim appeals and equipment orders, GPT-6 Astra scored 94.3 percent and the smaller GPT-6 Luna 91.8 percent at 8.5 cents per task, 98 percent less than Astra, HIT Consultant reports. In OpenAI's own evaluation, physicians rated 99 percent of 325 EHR inbox tasks completed by Astra as clinically safe.

The evidence has limits. The clinical safety figures come from OpenAI's evaluations, and no independent evaluation of these agents in a live health system has been published. A benchmark score of 91.8 percent also means a share of tasks fall short, which matters for submissions that touch payers and patients.

The controls are where compliance, privacy and CIO teams should focus. Organizations decide which applications the agents can reach and which steps require human review, and every action is recorded in audit logs, per Becker's. Ambience demonstrated an agent that stops to ask a clinician when a record is ambiguous and waits for approval before submitting a prior authorization. An agent that clicks through EHRs and payer portals is, in effect, a new privileged user, and access should be scoped accordingly.

The move: pick one bounded back-office queue, such as prior authorization submissions, and define the audit, access and human sign-off rules an agent would need before any pilot. Measure backlog, turnaround and error rates against today's baseline so the return to staff and patients is visible.

Do this week. Pick one bounded back-office queue, such as prior authorization submissions, and define the audit, access and human sign-off rules an agent would need before any pilot.

Run this

The prompt to act on this item, for your own model.

Paste it into Claude, ChatGPT or Copilot and fill the bracketed slots with your own numbers. The output is a draft for your team to check, built only from the sources on this page. Never paste patient data. How these prompts are written.

You are a compliance, privacy and security advisor to the executive team of [your organization], a [type of health organization] that uses [your EHR] and submits work through [your main payer portals].

Context: Becker's Hospital Review reports that OpenAI is building computer-use agents with CVS Health, Cigna Healthcare, Abridge and Ambience Healthcare to handle referrals, prior authorizations, discharge planning and other work across existing healthcare software. The agents, announced Nov. 16, read screens, click, type and move between applications such as EHRs and payer portals, and health systems can access them through ChatGPT. HIT Consultant reports the release also adds 20+ integrated health tech plugins, and that OpenAI cites clinical safety benchmarks for its models. Those benchmarks are OpenAI's own evaluations as reported by HIT Consultant; no independent evaluation of the agents in a live health system exists yet. OpenAI News describes the release in its own announcement.

Task: Treat an agent of this kind as a new privileged user. For one bounded back-office queue, [the queue you choose, such as prior authorization submissions], define the audit, access and human sign-off rules the agent would need before any pilot. Use our baseline: [current monthly volume in that queue], [current turnaround time], [current error or denial rate], and [current staff hours spent on the queue].

Output: a draft control workflow for the executive team, with these sections: queue scope and exclusions; systems the agent may and may not reach; access level and credentials; steps that require human review and sign-off before submission; audit log requirements and who reviews them; baseline metrics to compare in a pilot; open questions for the vendor. Assign an owner to each control from [names or roles of your owners].

Guardrails: Use only the facts above and the slots I have filled. Cite Becker's Hospital Review, HIT Consultant and OpenAI News by name where you rely on them. Flag anything uncertain or that must be verified with the vendor or our own teams. Do not request or include patient data. Nothing here is medical advice. Label the result as a draft for the team to check.

Open in ClaudeOpen in ChatGPT

The claim.
OpenAI says the agents work across healthcare software to get real work done, and HIT Consultant reports OpenAI cites clinical safety benchmarks for its models.
The evidence.
The benchmarks are OpenAI's own evaluations as reported by HIT Consultant; no independent evaluation of the agents in a live health system exists yet.

By seat

For the Chief AI Officer. Decide whether screen-level agents fit your governance model and require logging of every click an agent makes in an EHR or payer portal.

For the CFO and revenue cycle. Size the prior authorization and referral labor an agent pilot could take on and set the measure, such as turnaround or touches per case, before signing.

For the CIO and CTO. Inventory which EHR and payer portal sessions an agent would need credentials for and how those accounts are provisioned and revoked.

For the Compliance, privacy and legal. Determine how agent actions on PHI are attributed, logged and covered under existing business associate terms.

Disclosure at publication: OpenAI is one of the enterprise AI platforms Ecofy implements for clients.

Checked before publishing. 3 independent outlets. 1 figure matched to the source text.

Sources

  1. Becker's Hospital Review, OpenAI builds healthcare AI agents with CVS, Cigna, Abridge, Ambience
  2. HIT Consultant, HLTH26: OpenAI Expands ChatGPT for Healthcare: Computer-Use Agents, 99% Clinical Safety Benchmarks, and 20+ Integrated Health Tech Plugins
  3. OpenAI News, More ways to move care forward with ChatGPT for Healthcare

Drafted by Lumen from the sources above, scored by a second pass on six criteria, then checked by rule against those sources (every figure, two independent outlets or a primary source) and published only when it passes. A person can pull any item, and corrections are published here. Apolitical, always. Nothing here is medical advice. How Your Chief AI Officer is made.

Also in this brief

Your Chief AI Officer

What AI and robotics can do for your organization today, and what is coming in thirty days.

Every weekday morning, ordered for your role. Free.

One click to confirm, one click to leave, and you can change your role or cadence any time. Your address stays with Ecofy and is never shared with partners.