Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days

OpenAI Puts Computer-Use Agents on Hospital Admin Work: CVS Health Cleared a Months-Long Backlog in One Weekend

OpenAI Puts Computer-Use Agents on Hospital Admin Work: CVS Health Cleared a Months-Long Backlog in One Weekend

OpenAI says its computer-use agents are now available for healthcare teams through ChatGPT for Healthcare, ChatGPT Work and Codex. CVS Health, Abridge, Cigna and Ambience are deploying them, and Stanford's HealthAdminBench scores GPT-6 Astra at 94.3 percent while GPT-6 Luna scores 91.8 percent at about 8.5 cents per task. All vendor-reported figures.

OpenAI has opened its computer-use agents to healthcare teams, letting models that read the screen, click, type and move across applications handle the multi-step administrative work that clinics and insurers still do by hand. The capability is available from today in ChatGPT for Healthcare, ChatGPT Work and Codex, and the company published deployment cases from four healthcare organizations along with evaluation numbers that are worth reading with the usual vendor-reported caveat.

The flagship customer story comes from CVS Health, where agents navigate existing clinical applications to fill gaps that APIs do not cover. One production agent cleared a months-long backlog of document-sending tasks over a single weekend, and another now works through the system-level steps behind tens of thousands of referrals per day. CVS said patient data stays inside its own infrastructure and is not used to train OpenAI's foundation models, and the company open-sourced an Agent Kit built on OpenAI's Agents SDK.

The other three deployments follow the same pattern of putting screen-operating agents into workflows that were never designed for automation. Abridge uses agents to coordinate discharge barriers identified from clinical conversations, working in parallel on medication access and home-health scheduling. Cigna Healthcare says agents can now pull together benefits, coverage, provider and appointment context from more than ten separate systems in seconds; its CIO Madhu Nutakki said the program moved from concept to deployment in under eight weeks. Ambience Healthcare pre-fills prior-authorization requests on insurer portals using evidence from patient records, and pauses to ask a clinician when a record is ambiguous.

On evaluation numbers, OpenAI reports that in 325 audited tasks on an EHR inbox workload, 99 percent were rated clinically safe, and across 646 helpfulness evaluations doctors chose the top two ratings 93 percent of the time on a five-point scale. On Stanford's independent HealthAdminBench benchmark at maximum reasoning effort, GPT-6 Astra scored 94.3 percent while the cheaper GPT-6 Luna scored 91.8 percent at roughly 8.5 cents per task, a 98 percent cost reduction. These are the company's own reported figures and have not been independently verified.

Access is relatively broad for a healthcare product: the agents run through the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians, which is free for US clinicians to sign up for. OpenAI says HIPAA-eligible workflows are supported under business associate agreements, with role-based access controls, audit logs, configurable approval steps and scoped agent permissions. OpenAI also named five health startups already building on the stack: Assort, Bunkerhill, Forus, Gravity and Tennr. Assort, for example, uses Codex computer-use to complete referral-closure steps inside the Athena EHR interface that its API does not cover.

The significance is less about benchmark points than about where the work happens. Healthcare administration is one of the largest non-clinical cost centers in the US system, and it runs on legacy interfaces that resist API integration, which is precisely the environment where computer-use agents have an edge over traditional integration projects. The open question, as always in this sector, is whether regulators and hospital IT departments accept an agent clicking through patient-adjacent systems; OpenAI's answer, for now, is audit logs, human approval gates and a contractual promise that the data never becomes training fuel.

Comments (0)

Log in to join the discussion

Log In

No comments yet