The Current

OpenAI Details Computer-Use Agents for Healthcare Administrative Work

The company reports benchmark results and names pilot partners including CVS Health, Cigna, Abridge and Ambience.

useful business · for technical · November 17, 2026

OpenAI announced on November 16, 2026, new results and partnerships for 'computer-use agents' — AI that completes multi-step tasks by clicking, typing and moving between applications in software teams already use. According to OpenAI, these capabilities are available through the OpenAI API, ChatGPT for Healthcare, and ChatGPT for Clinicians, with support for HIPAA-compliant workflows, configurable approvals, agent access controls and audit logs. OpenAI described a demo built on synthetic patient data in which an agent reviews records in an EHR, drafts an MRI order, an encounter note and a patient reply for clinician review, then prepares a prior authorization request in an insurer's portal. The company reported that in clinical inbox tasks, 99% of 325 reviewed tasks were rated safe, and physicians rated the work 'very helpful' or 'extremely helpful' in 93% of 646 assessments. On HealthAdminBench, described as an independent benchmark from Stanford researchers, OpenAI said GPT-6 Astra earned 94.3% of rubric points and GPT-6 Luna earned 91.8% at 8.5 cents per task. OpenAI stated CVS Health used an agent to clear a months-long document-sending backlog over a weekend and is open-sourcing an 'Agent Kit.' These figures come from OpenAI's own announcement.

  • Available via OpenAI API, ChatGPT for Healthcare, and ChatGPT for Clinicians; supports HIPAA-compliant workflows and audit logs
  • OpenAI reports 99% of 325 clinical tasks rated safe; 93% of 646 helpfulness assessments rated very/extremely helpful
  • On HealthAdminBench, GPT-6 Astra scored 94.3%, GPT-6 Luna 91.8% at 8.5 cents per task
  • Named partners include CVS Health, Cigna, Abridge and Ambience; CVS is open-sourcing an Agent Kit

What it means for you

This is a vendor announcement showing AI agents can now operate ordinary software screens — clicking and typing through portals and record systems — to handle repetitive administrative steps like prior authorizations and referrals. The headline numbers (99% safe, 94% benchmark) come from OpenAI itself, so treat them as claims, not independent proof. The genuinely interesting signal is the approach: agents that navigate the applications you already use, for tasks that never had a clean integration.

Try this

Pick one repetitive, screen-bound task your team does across multiple systems — pulling data from one portal and re-entering it in another — and map it step by step. That map tells you whether the real problem is the manual work or a broken handoff upstream, and you need it before any agent could help.

Who should care

Operations and revenue-cycle teams in healthcare organizations, and anyone who owns high-volume, cross-system administrative workflows like authorizations, referrals or claims. Also IT leaders who already run HIPAA-governed environments.

Skip this if

You're not in healthcare, or your administrative work already runs through stable integrations. For everyone else, this is early-stage and partner-specific — worth watching, not acting on this week. And keep a person reviewing anything that touches patient care, money or a clinical decision.

Sources: OpenAI — read the original

Keep reading

The rest of this story is free with your email.

One field, asked once. It unlocks every story on The Current in this browser, including what it means for you and what to try.

We’ll email occasionally; decline any time. Privacy

← All stories