Subscribe
AI Agent Use Cases

OpenAI's computer-use agents are now working inside hospitals and health plans

Tuesday, November 17, 2026 · 4 min read

OpenAI announced Nov 16 that computer-use agents are now available for healthcare teams, operating existing hospital and payer software to carry out multi-step work — and CVS Health is already using them in production.

Bullish$CVS$CI

📊 Scorecard: our Bullish call on $CVS is scored from $88.59, the last close before it ran. Its first result comes after the next close.

OpenAI's computer-use agents are now working inside hospitals and health plans

The News

OpenAI announced on November 16 that computer-use agents are now available for healthcare teams through ChatGPT for Healthcare, ChatGPT for Clinicians and the OpenAI API, alongside more than 20 new healthcare plugins and improved clinical-evidence search. The agents operate existing EHR and payer-portal software — reading screens, clicking and typing — to carry out multi-step work such as prior authorizations, referrals, discharge planning and specialist-consult support.

OpenAI reports physicians rated 99% of 325 reviewed GPT-6 Astra EHR-inbox tasks as clinically safe, and 93% of 646 assessments as very or extremely helpful. On Stanford's independent HealthAdminBench, GPT-6 Astra scored 94.3% and the cheaper GPT-6 Luna scored 91.8% at about 8.5 cents per task. All of these figures are OpenAI-reported, and the safety scores are evaluations, not clinical outcomes from live patient care.

CVS Health is using the agents in production — one cleared a months-long document backlog over a weekend — and Cigna has moved its deployment from concept to production in under eight weeks, while Abridge and Ambience are piloting or developing with them. That is the difference between a demo and a deployment: major health companies are already running the agents on real work.

Healthcare's paperwork is the largest untapped workload in the economy — and an agent that can click through it is worth more than a chatbot.

Why It Matters

Healthcare administration is the largest paperwork burden in the economy. Prior authorizations, referrals, discharge planning and specialist consults are multi-step jobs that eat hours of staff time and delay care for patients. An agent that can operate the existing software — reading screens, clicking, typing — attacks that backlog without asking hospitals to rip out and replace systems that cost billions to install.

The economics are what make this real. At about 8.5 cents a task for the cheaper model, and with the flagship model scoring 94.3% on Stanford's independent benchmark, the cost of trying the technology is trivial next to the cost of the staff time it replaces. That is why a company like CVS, which runs pharmacies, clinics and a health plan, would put it into production rather than run another pilot.

The risk is also plain. Healthcare is the one industry where an agent's mistake can hurt a real person, and 'clinically safe' in a review is not the same as safe in a live clinic. OpenAI's numbers are evaluations, and the guardrails — human oversight, audit trails, escalation — will decide whether this scales or stalls after the first public error.

Analyst's Take

Our call is Bullish. This is the first time a frontier AI company has put computer-use agents into live healthcare operations with a named customer, and the numbers are strong: 99% of reviewed tasks rated clinically safe, 94.3% on Stanford's independent benchmark, at about 8.5 cents a task. A months-long backlog cleared over a weekend at CVS is the kind of result that sells itself to every other health plan watching.

The caveat is the one OpenAI itself would insist on: evaluations are not outcomes. A physician rating a task safe in review is not the same as a patient being safe in a live clinic, and the 99% figure comes from OpenAI's own reporting. The real test is what happens when these agents run for months across thousands of cases, with errors counted in the open.

For readers, the signal is the deployment, not the demo. CVS in production, Cigna piloting, and a per-task cost under a dime add up to a product health systems can actually buy. Watch for the next customer names and for any safety incident — the first one will set the tone for the whole category. Reader poll: Would you trust an AI with your paperwork? Vote: Yes, admin only, With a human check, Not yet.

Are you bullish or bearish on this story?

By The Numbers

99%Of 325 reviewed GPT-6 Astra EHR-inbox tasks rated clinically safe
93%Of 646 assessments rated very or extremely helpful
94.3%GPT-6 Astra score on Stanford's HealthAdminBench
91.8%GPT-6 Luna score, at ~8.5 cents per task
20+New healthcare plugins launched alongside the agents
1Months-long document backlog cleared over a weekend at CVS

FAQ

What is a computer-use agent?

It is an AI that operates software the way a person does — reading screens, clicking buttons and typing — instead of needing a special connection to each system. That lets it work inside existing hospital and insurance software without those systems being rebuilt.

What is an EHR?

An EHR is an electronic health record — the software where hospitals store patient charts, test results, prescriptions and visit notes. It is the main system doctors and nurses work in all day.

Are the safety results proven in real clinics?

Not yet. The 99% clinically-safe rating comes from physicians reviewing 325 tasks, and the 94.3% score comes from a Stanford benchmark — both are evaluations, not clinical outcomes from live patient care.

👁

Source: Unite.AI

Get the daily briefing. AI news with a markets lens — tickers, bullish/bearish reads, zero hype. Subscribe free on Substack

More on $CVS

All $CVS coverage →

More from AI Agent Update