OpenAI has introduced computer-use agents for US healthcare workflows. CVS Health and Cigna are already using them, and the safety evidence so far is OpenAI's own.

OpenAI has brought its computer-use agents into US healthcare, letting software click and type its way through clinical systems such as electronic health records and insurance portals instead of connecting through APIs

1

2

. The pitch is aimed at the paperwork that sits between doctors and patients.

Source: Interesting Engineering

Source: Interesting Engineering

Why healthcare

US healthcare staff spend much of their day moving information between systems that do not talk to each other, and that paperwork is what the agents are built to take on

1

. OpenAI frames the goal as cutting administrative workload while keeping clinicians in the decisions that matter

1

.

The evidence so far

In OpenAI's clinical safety evaluation, physicians reviewed 325 tasks that GPT-6 Astra performed on electronic health record inboxes and rated 99 percent clinically safe

1

2

. Across 646 assessments, 93 percent were rated very or extremely helpful

1

2

.

On HealthAdminBench, a benchmark built by Stanford researchers, Astra scored 94.3 percent

1

2

. Luna scored 91.8

1

2

.

OpenAI also published cost figures. It says one of the two models completed benchmark tasks at about 8.5 cents each, 98 percent less than the other

1

2

. The two reports disagree on which model that was

1

2

.

Who is already using it

CVS Health, a major US health company, uses the agents to move documents, handle referrals and support specialist consultations

1

2

. Per OpenAI's account, one agent cleared a document backlog built up over several months in a single weekend

1

2

. Another handles software steps for tens of thousands of consultations daily

1

.

Cigna Healthcare uses them to pull coverage, provider and appointment details from more than 10 separate systems in seconds

1

2

. Abridge is building agents for inpatient care coordination such as medication access

1

. Ambience uses the technology to prepare insurance authorization requests, pausing for clinician input when records are ambiguous

1

.

In a demonstration using synthetic patient data, an agent handled a magnetic resonance imaging request for a patient with lower back pain

1

2

. It reviewed the record, therapy and imaging history, then drafted an imaging order, a clinical note and a prior authorization request for clinicians to check before anything happened

1

2

.

Getting access, and staying compliant

OpenAI offers the agents through its API, ChatGPT for Healthcare and ChatGPT for Clinicians

1

2

. Organizations can set approval requirements, restrict what agents may access and keep audit logs for deployments that meet US health-privacy rules

1

2

. OpenAI says it does not use organizational data to train its foundation models

1

2

. It says its systems also capture screenshots and action records so staff can verify what the agents actually did

1

.

Source: Gadgets 360

Source: Gadgets 360

Both sets of results are figures OpenAI reported from its own evaluation; the sources provide no independent verification

1

2

.

© 2026 TheOutpost.AI All rights reserved