KevinMD.com: Social media’s leading physician voice
The KevinMD newsletterReal physician voices, twice a weekFree, and one click to unsubscribe.
Subscribe free →

Health technology

Harvey Castro, MD, MBA, on AI in medicine: a KevinMD subscriber spotlight

Harvey Castro, MD, MBA

Harvey Castro, MD, MBA is an emergency physician, chief AI officer, and physician futurist known as DR GPT™. He has published 98 essays on KevinMD since January 2023 and has been a guest on The Podcast by KevinMD three times. Most of that work is about the same two questions: what medicine should let a machine do, and who is responsible when it does.

This spotlight is part of a series on subscribers to the KevinMD enhanced author page.

He published 59 essays in 2023, 18 in 2024, 11 in 2025, and 10 so far in 2026. The 2023 pieces cover ChatGPT’s first year in medicine as it happened, including privacy, hallucinations, and whether the tool could ease the physician and nursing shortage. In 2024 and 2025 he wrote more about hardware and standards, including Apple Vision Pro, ambient sensors, and the protocol that lets a model read a chart. The 2026 essays are aimed at people who make decisions: who is liable for an AI-drafted note, what a hospital board gets wrong, and what a physician should do before trusting any output. He has also written regularly about burnout, resilience, leadership, and what a physician’s job becomes once a machine handles recall.

Harvey Castro at a glance

98essays on KevinMD

3podcast episodes

January 2023writing on KevinMD since

12questions answered below

Will AI replace physicians?

No, according to Harvey Castro, and he thinks the question distracts from a better one. In “Clinical AI should replace workflows, not physicians,” September 2026, he writes that “the most useful first wave of clinical AI will not replace physician judgment.” The work he wants automated is the administrative work that should never have been assigned to physicians. His answer sits among many others on the same question in the KevinMD record on artificial intelligence.

Who is liable when an AI drafts a clinical note and a physician signs it?

The physician, and Castro says the profession has not paid attention to the contract language that makes it so. “AI in clinical documentation: Who is liable for medical errors?,” February 2026, opens with an attending signing a note he did not write. Vendors, Castro observes, assign ownership of every output to the user and disclaim the rest, which leaves the clinician responsible for a document that also serves as a billing instrument and as evidence in a malpractice claim. His conclusion: “When AI-generated reasoning meaningfully shapes care, authorship determines accountability.” Accountability, he writes, does not disappear because the reasoning came from a system that cannot hold a license or testify in court. The same gap runs through the KevinMD record on medical errors.

What mistakes do hospital boards make with AI strategy?

The most common one, in Castro’s account, is approving tools before deciding what problem they are meant to solve. In “AI strategy for hospitals: the 5 mistakes boards make,” September 2026, he writes that in many organizations “AI is entering through the side door before leadership has created a clear strategy.” He recommends that boards begin with the problem and that no initiative be approved without input from the clinicians who will have to use it.

How should physicians use AI responsibly?

Castro’s first rule is to “treat AI output as a draft, not a decision,” from “7 rules for how physicians should use AI responsibly,” September 2026, and the other six follow from it. He expects verification, more than recall, to be the skill that matters in the next decade, and he writes that a tool should make clinicians better informed without making them less responsible.

What safety protocols should a hospital have before deploying ChatGPT or another large language model?

Castro lists five, all of them required before go-live. “ChatGPT Health in hospitals: 5 essential safety protocols,” January 2026, lays them out: encryption and access control, a human in the loop, simulated testing in a sandbox, bias audits, and plain disclosure to patients. Without them, he writes, hospitals risk “misinformation, privacy breaches, and erosion of patient trust.”

What is agentic AI in medicine?

Not a better chatbot, Castro writes, but a system that acts on its own within limits, which changes what supervising it means. In “Agentic AI in medicine: Moving beyond ChatGPT,” January 2026, he describes systems that “break down complex tasks into intermediate steps and execute them autonomously” inside set limits, and that keep a patient’s context from one step to the next instead of starting over with each prompt. Checking an agent’s work, he argues, is not the same task as checking a paragraph of text.

Why does clinical AI underperform at the bedside, and what is the model context protocol?

Because the model cannot see the chart, Castro argues, and the protocol is one answer to that. “Model context protocol: the standard that brings AI into clinical workflow,” May 2025, opens with the claim that “the real limitation of clinical AI isn’t the model; it’s the context.” His examples are the blood pressure entered three minutes ago and the allergy buried in yesterday’s note, neither of which the model can see. The essay then explains a standard that lets an agent reach clinical data without a custom build for every vendor. His documentation writing sits next to the KevinMD record on the electronic health record.

Can a large language model be HIPAA compliant?

Not by default, and Castro treats that gap as the main obstacle to clinical use. In “Why Grok 4 could be the next leap for HIPAA-compliant clinical AI,” July 2025, he writes that most models still sit outside the perimeter that protects patient data, and that the heavy lift of getting one inside falls on hospitals, not vendors. His conclusion is addressed to physicians: “Clinicians who master HIPAA-compliant AI will replace those who don’t.”

Is it safe to paste patient or company data into ChatGPT?

No, because the data can end up outside the organization’s control, and Castro made that point in April 2023, before most health systems had a policy. “The unintended consequences of feeding company data into ChatGPT: a guide to safeguarding your intellectual property” used the Samsung leak to warn that users “might unintentionally expose their proprietary research data to a third party.” He had made the compliance version of the argument two months earlier, in “AI in health care: Meeting HIPAA standards with ChatGPT,” February 2023.

How common is physician burnout, and what helps?

Common enough, Castro writes, to be a systems problem rather than an individual one. In “The invisible epidemic: How physician burnout is affecting the medical community,” April 2023, he cites studies showing “approximately 50 percent of physicians experience symptoms of burnout,” and connects it to patient care and medical errors. Three years later his emphasis had shifted. “Physician resilience: Why systems matter more than heroism,” February 2026, argues that the physicians who last over a career rely on “a well-built internal operating system” instead of willpower. Both essays belong to the territory of the KevinMD record on physician burnout.

Can AI fix the physician and nursing shortage?

Castro’s answer is that it can reduce the load on the people who remain, but it does not produce more of them. “Tackling the health care crisis with artificial intelligence: Combating physician and nursing shortages in the United States,” March 2023, cites the Association of American Medical Colleges projection of “a shortage of 54,100 and 139,000 physicians by 2033,” and separates the causes technology can affect, such as administrative burden and burnout, from the ones it cannot, such as an aging population and a workforce nearing retirement. The staffing side of that argument continues in the KevinMD record on nursing.

What makes a physician necessary in the age of AI?

Judgment in situations no model has been trained on, in Castro’s telling. “In the age of AI, what makes a physician REAL?,” February 2026, opens at 3:17 a.m. in a trauma bay with a hypotensive patient, no CT yet, and no algorithm offering an answer. Castro argues that as the tools improve, physicians have to become “less defined by recall, more defined by judgment,” and that no algorithm resolves the case where a treatment with a statistical benefit conflicts with what the patient values.

Real physician voices, twice a week

Free, and one click to unsubscribe.

No spam. We never share your email.

Questions on this page · All spotlights

Castro has been on The Podcast by KevinMD three times: on the physician and nursing shortage in April 2023, on Apple Vision Pro and diagnostics in May 2024, and on hospital AI guardrails in March 2026. Asked in the last of those for a take-home message, he told listeners to “think of AI like riding a bike.” The fear at the start is normal, he said, and the only way past it is to use the tool.

About the KevinMD enhanced author page: This spotlight exists because Castro’s author page is an enhanced author page: his photo, biography, affiliations, books, and links on one page that ranks for his name and that ChatGPT, Perplexity, and Google’s AI read when someone asks who he is. The same branded attribution appears at the end of every article he has published, so each essay listed below points back to a page he controls. Subscribers get three things the free track does not:

Publish in 1 to 3 days. Free-track articles wait 5+ weeks, and the wait keeps growing. Subscribers skip the line.

Get a guaranteed podcast slot. Free slots fill within hours and the waiting list runs into the hundreds. Subscribers are guaranteed a spot.

Control your Google and AI identity. Your KevinMD author page is one of the highest-ranking results for your name. Make it represent you, in your words.

Enhanced author page

Upgrade your author page

The enhanced author page is $99/month or $999/year.

Cancel anytime for any reason, and your author page will revert to the standard attribution.

Upgrade your author page

Have questions before you subscribe? Contact Kevin.

Everything Harvey Castro has written for KevinMD:

Harvey Castro on The Podcast by KevinMD:

Share this articleLinkedInFacebookXEmail

Tagged as: Health IT and AI in Medicine

Real physician voices, twice a week

Free, and one click to unsubscribe.

No spam. We never share your email.

More by Subscriber Spotlight by KevinMD

Keep reading

More in Health Technology →

Leave a Comment
KevinMD.comHealth technologyHarvey Castro, MD, MBA, on AI in medicine: a KevinMD subscriber spotlight
0
0