Why Emergency Physicians Make the Best AI Governance Leaders

A 2026 case for putting operational ED physicians on every clinical AI procurement committee — and giving them veto authority.

Don’t forget to include an emergency medicine phyisican on your AI governance committee.

The hospital’s AI governance committee met on a Tuesday. Eighteen people in the room. Two attorneys, three IT directors, the CMIO, the compliance officer, two members of the board, the chief nursing officer, four department chairs, the head of quality, the chief financial officer, and one emergency physician. The agenda was the procurement of an enterprise ambient documentation AI tool. The discussion ran ninety minutes. The emergency physician spoke twice. Both times the discussion changed direction. The committee voted to delay the procurement and ask for the override-pathway documentation, the failure-mode analysis from the vendor’s prior deployments, and the contractual commitment around productivity creep before approval. The CFO was unhappy. The CMIO smiled.

That meeting is one example of a pattern I have watched repeat across health systems, military medical commands, and regional disaster coalitions for the last three years. When AI governance committees include an emergency physician — a real one, with operational authority and the temperament to use it — the decisions get sharper, the procurement gets harder for the vendor, and the patient-safety lens stops being a slide at the end of the deck. When they do not include an emergency physician, the committee converges on whatever the vendor brings to the room.

This is not a flattering observation about emergency medicine and it is not a criticism of the other disciplines on those committees. It is a structural argument. Emergency physicians are operationally trained to do exactly the work that good AI governance requires, and most institutions have not yet recognized the fit.

The Job Description Already Matches

The National Institute of Standards and Technology AI Risk Management Framework released in January 2023 organizes AI governance around four functions: govern, map, measure, and manage. Read those functions through an emergency physician’s lens and the alignment is immediate.

1. Govern is a culture-and-process function. The ED runs on culture and process.
2. Map is contextual risk identification. Emergency physicians map context under time pressure on every patient.
3. Measure is the iterative assessment of system performance against defined metrics. The ED is the most heavily measured environment in clinical medicine.
4. Manage is the operational allocation of risk-mitigation resources in real time. That is the entire job of the emergency physician on a busy shift.

The four NIST functions are not a metaphor for emergency medicine work. They describe the same cognitive activity. The institutions that have already figured this out are quietly putting emergency physicians on their AI governance committees and watching the quality of decisions improve. The institutions that have not yet figured it out are still letting the vendor draft the slide deck.

The Specific Skills That Translate

There are five skills that emergency medicine training builds at a depth no other discipline matches at the same career stage, and all five are central to credible AI governance.

1. Operating under uncertainty with incomplete information. The emergency physician makes high-stakes decisions in the first two minutes of a patient encounter on roughly forty percent of the data that would be available at hour twelve. That is the same epistemic posture AI governance requires. The vendor will not give you the algorithm. The institution will not give you the historical performance data. The regulator will not give you the case law. You make the procurement decision anyway, on the evidence you can get, with the decision-margins documented in writing for the institution to argue against later.
2. Reading the room under pressure. The emergency physician reads families, consultants, charge nurses, and administrators in real time and adjusts the communication strategy without breaking the clinical narrative. AI governance committees are largely a communication problem. The vendor is selling. The CFO is calculating. The CMIO is hedging. The lawyer is risk-translating. The chief nursing officer is workforce-modeling. The emergency physician is one of the few participants who can hold all five frames at once and translate between them. That fluency is not optional for an effective AI governance chair.
3. Calling the bluff. Vendors selling AI tools to healthcare are reliably overconfident about three things: validation data, edge-case performance, and the regulatory pathway. Emergency physicians are professionally calibrated against overconfidence — every shift is a tutorial in the gap between what we think we know and what the patient actually has. Asking the vendor to produce subgroup performance, to document the failure modes from prior deployments, and to commit in writing on the FDA Predetermined Change Control Plan they have filed is not adversarial. It is the same diagnostic discipline we use on a febrile infant. The vendor is not used to it. The vendor needs to get used to it.
4. The override question. The single most important governance question for any clinical AI system is what happens when the algorithm is wrong and the clinician knows it. The answer to that question — the design of the override pathway, the documentation standard for divergence, the post-event review structure — is the operational core of clinical AI governance, and emergency physicians have been thinking about override and decision-margin protocols for a generation. For a fuller treatment of how this plays out in mass casualty and disaster settings, see AI in Mass Casualty Triage: Promise, Peril, and Override on the GMOC blog.
5. Systems thinking under load. The ED is the only clinical environment in the hospital where a single physician is simultaneously responsible for system throughput, clinical decision-making on individual patients, supervision of trainees, and coordination with external partners — EMS, consultants, transfer centers, the inpatient hospital — over a multi-hour continuous block of time. That is precisely the cognitive load profile of an AI governance role at a health system scale. The board chair has the authority and the lawyer has the framework, but the emergency physician has the wired reflex for managing the whole system without dropping any single piece of it.

What the Evidence Base Actually Supports

The argument is not just temperamental. The 2026 evidence base on AI governance failure modes maps onto the gaps that emergency physicians are uniquely positioned to fill.

The U.S. Food and Drug Administration’s finalized guidance on Predetermined Change Control Plans, published in December 2024 and being operationalized through 2026, places the burden on the developer to specify in advance which model modifications will be allowed without a new marketing submission, the methodology by which those modifications will be implemented, and the impact assessment for safety and effectiveness. For health system AI governance, the implication is that the committee evaluating a clinical AI tool must be able to read a PCCP, ask sharp questions about its scope, and refuse the procurement if the answers do not hold up. Read the FDA guidance directly: Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions.

The European Union’s AI Act, which entered into force on August 1, 2024, and reaches full applicability for high-risk AI systems on August 2, 2026, classifies most clinical AI tools as high-risk and imposes hard obligations on providers and deployers around data governance, risk management, post-market monitoring, and human oversight. For U.S. health systems with European exposure or European patients, those obligations are operational reality. See the European Commission’s overview at AI Act | Shaping Europe’s digital future.

The World Health Organization’s report on Ethics and governance of artificial intelligence for health grounds AI for health in human rights and patient autonomy, lays out six consensus principles, and provides the international anchor for institutional governance frameworks. Reading those three documents — FDA, EU AI Act, WHO — in sequence is the entry-level credential for a credible AI governance committee chair in 2026. The emergency physician who has read them, mapped them against the institution’s actual procurement decisions, and translated them into a one-page committee charter is operationally ready for the role.

What Institutions Should Do Now

1. Put at least one practicing emergency physician on the institutional AI governance committee. Not as a representative of the ED — as a governance participant in their own right. Choose someone with operational authority in the department, comfort with regulatory language, and the temperament to ask the vendor uncomfortable questions in a room full of finance and IT colleagues.
2. Compensate the role. AI governance is not committee work. It is a regulatory, clinical, and operational function with institutional liability attached. Build the protected time and the financial accountability into the role before you fill it, or you will get the level of engagement you pay for.
3. Give the role veto authority on clinical AI procurement that touches emergency department workflows. The emergency physician is the only person in the room whose patients will be the first to feel the failure mode of a poorly governed clinical AI tool. The procurement structure should reflect that asymmetry.
4. And insist that the committee read, in original, the FDA PCCP guidance, the EU AI Act high-risk obligations, the WHO ethics framework, and the NIST AI RMF before approving any enterprise clinical AI procurement. Vendors will summarize those documents for you. The summaries will be wrong in ways that favor the vendor. Read the originals.

Dr. Chet’s Take

I have served as the medical director of a telemedicine network, a state surgeon for a National Guard force, the medical lead for an air medical and critical care transport program, and the founder of a clinical operations consultancy. The pattern across all four roles is identical. The institutions that govern technology well are the ones that put operational physicians in the governance seat early, give them the authority to slow a procurement when the evidence does not hold up, and back them when the vendor escalates to the board. The institutions that govern technology badly are the ones that treat the operational physician as a stakeholder to be informed rather than a decision-maker to be empowered.

Emergency medicine produces the operational physician profile at depth and at scale, because the discipline trains the exact cognitive posture that AI governance demands. The institutions that recognize the fit are getting better procurement decisions, sharper override protocols, and more defensible documentation. The institutions that do not are getting whatever the vendor brings to the room. The choice is not subtle, and it is not difficult. It just has to be made.

AI Won’t Wait. Neither Should You.

If your institution is approving enterprise clinical AI procurement without an emergency physician at the governance table with operational authority, you are exposed. Consider enrolling in my course: AI in Emergency Medicine: Becoming AI Bulletproof. The course walks through the procurement checklist for clinical AI vendors, the FDA PCCP review framework, the EU AI Act high-risk obligations as they translate into U.S. institutional policy, the one-page AI governance committee charter we use in our regional consortium, and the override-pathway documentation standard that anchors the whole framework.

If you’re an emergency physician (or any clinician treating patients daily) trying to understand how AI will actually impact your clinical practice — not just the hype — I put together a free practical guide. You can download it here: AI in EM Survival Guide.

— — — — — — — — — — — — — — — — — — — —

Chester “Chet” Shermer, MD, FACEP is a Professor of Emergency Medicine, TeleHealth, HEMS and Critical Care Transport, and State Surgeon for the Army National Guard. He is the founder of Global MedOps Command and the creator of AI in Emergency Medicine: Becoming AI Bulletproof.

Read more on the GMOC blog at globalmedopscommand.com/blog.