OpenAI put computer-use agents into ChatGPT for Healthcare: they read the screen and click through EHRs and payer portals; CVS says one cleared a months-long document backlog in a weekend.
HHS finalized a HIPAA rule: and a patientâs chosen app is not your business associate.
Kaiserâs Northern California therapists ratified a contract with AI language: 97% yes, input on new technology, a no-layoff guarantee for nearly the full term.
CMS proposed a 545-page anti-fraud rule called CRUSH: every enrollment change reported within 30 days.
Google surveyed 250+ health executives; 30% have scaled agentic AI: 40% paused a project because the bill was unpredictable.
đ§ Podcast: The Gist Healthcare â What the new Congress means for hospitals and health systems: Vizientâs Shoshana Krilow expects prior-auth reform and permanent telehealth to be the bipartisan wins, and site-neutral payment the bipartisan risk.
đ©» Real or AI?
Two photos of a palm with red spots, both labeled âJaneway lesions.â One is real. The other was generated by AI. Which is which?
A
B
Answer at the bottom.
đ§ The AI Curbside â AI answers your questions
This is what AI thinks â not medical advice, not my opinion or my employerâs. Verify before you act.
âA colleague says she can e-prescribe from ChatGPT now. Really?â
Short answer: Yes, for non-controlled drugs, if sheâs a verified U.S. clinician.
What changed: Photon launched prescribing inside ChatGPT for Clinicians and ChatGPT for Healthcare on Monday. The tool drafts the script and screens interactions; the clinician reviews and authorizes each one. Itâs free through Dec. 31. Pricing after that isnât stated.
AIâs take: The release says records sync with Photon. It doesnât mention the EHR. A prescription that never reaches the chartâs med list is a reconciliation problem at the next visit.
đ€ âItâs just e-prescribing.â In a chat window, outside the chart.
đŹ The Big Thing
The agent passed 99% of 325 inbox tasks. Who gets the other three?
OpenAI announced Monday that its computer-use agents now work in healthcare: they look at the screen, then click and type through EHRs and payer portals.
Theyâre in the API, ChatGPT for Healthcare and the free ChatGPT for Clinicians. Whether thatâs preview or general availability, the announcement coverage doesnât say.
The numbers are OpenAIâs own. Reviewers rated 99% of 325 EHR-inbox tasks clinically safe, and physicians called the work âveryâ or âextremelyâ helpful in 93% of 646 assessments. The cheaper model reportedly runs about 8.5 cents a task.
CVS Health, Cigna, Abridge and Ambience are the named partners (Abridgeâs use is still in development). Ambience showed prior-auth submission on an insurer portal, with every click logged.
No API needed. If a person can click it, the agent can. That includes the wrong chart.
The design is careful on paper: agents draft messages instead of sending them, leave notes unsigned, and pause when the chart is ambiguous.
đŒ What it means for your job: the MA working the inbox and the prior-auth nurse on the portal are the first workflows in range. Their job shifts from doing the task to checking the draft, which is harder to staff and harder to notice when it slips.
đ History rhymes: When hospitals moved to computerized order entry, a 2005 JAMA study at one teaching hospital counted 22 new ways the system made medication errors easier. The authorsâ advice: plan for continuous revision, because every change creates new error risks.
đ€ â99% safe is better than my inbox coverage on a holiday weekend.â Maybe. But 325 tasks is one busy clinicâs Tuesday, OpenAI doesnât say who the graders were, and it has described the misses only in general terms: an agent sometimes acting without the proper sign-off.
đ€ âA human reviews everything.â The business case is that the human reviews less.
đŁïž Say it in the meeting: âShow me the 1%. What did the unsafe inbox tasks look like, and who here would have caught them?â
The first clinicians to bargain over the algorithm
About 2,400 Kaiser therapists, social workers and psychologists in Northern California ratified a four-year contract with 97% support, their union said Monday.
Per the union, it gives clinicians input on new technology, the right to bargain over its effects, and a clinician advisory group Kaiser must consult before changing the model of care.
A separate state complaint over Kaiserâs use of algorithms to triage patients is still open.
âHealthcare needs a chief token officerâ
Thatâs Aashima Gupta, Google Cloudâs global director of healthcare, on a Google-commissioned survey of more than 250 health executives.
About 60% hit trouble moving AI from pilot to live systems. 40% paused something over unpredictable costs.
đ€ âA vendor survey that says buy more carefully from vendors.â Yes.
đ§Ș Evals Desk
Objectify, Donât Delete: Complaint-Derived Evaluation of Ambient Clinical Notes
The authors mined eval criteria from cliniciansâ complaints about scribe notes, then split each complaint into narrow yes/no checks. The checks found 11 planted defects at median recall 0.95. The same judge, asked the complaint as one broad question, also flagged 70% of clean notes.
Why it matters: The complaints your rubric canât grade are the ones clinicians raise most. Thatâs a rubric bug, not a model bug.
đĄ BTW: Photon CEO Otto Sipe spent nearly a decade in and around pharmacy at startups and Walmart before starting the company. He told Fortune that taking on the e-prescribing incumbents was âa crazy bet for people who work in health care.â
đ©» Real or AI answer: B is real, A is AI.
The tell: Very subtle. The fakeâs lesions are barely visible and spread evenly across the palm like a texture rather than clustering on the thenar and hypothenar eminences, and the palmar creases multiply into a fine crosshatch that doesnât follow real flexion lines.
The pearl: Janeway = painless, flat (septic microemboli). Osler = painful, raised (immune complex). Both are uncommon. Fever + new murmur + IV drug use or a prosthetic valve â 3 sets of blood cultures before antibiotics. Learn more: Endocarditis (MedlinePlus).
Proof: A was made with Nano Banana 2.1 on 2026-11-12. B is a published, de-identified teaching image, not patient data: âJaneway lesionâ by Warfieldian, CC BY-SA 4.0, Wikimedia Commons, resized, shared under the same license.
This post is almost entirely AI-generated. I run several agents that compile it for me every day, and I share it because I think others may find it useful too. It may contain errors, so please verify anything important for yourself.
The opinions here are mostly the AIâs. They are not medical advice, and they donât necessarily reflect my views or those of my employer. Still, I find it interesting to see in one place whatâs happening in informatics and health tech, and how people might be reacting to the news. I hope you find it of value as well in this fast paced world that is accelerating.
â Kevin & Fable 5.2




