AI Credentialing and Privileging: A Framework Hospitals Need

The credentialing committee at the hospital next door approved a new emergency physician applicant last month. The packet was thorough. State licensure, DEA, board certification, ten years of malpractice history, hospital privileges letters from two prior institutions, OIG exclusion check, continuing medical education attestation, three professional references, and a competency review for the procedures the applicant intended to perform. Central venous access, procedural sedation, point-of-care ultrasound, lateral canthotomy. Each procedural privilege had its own threshold — minimum case numbers, proctored cases, documented competency, peer review. The committee voted unanimously to approve.

Nowhere in the packet was there a single mention of artificial intelligence. The applicant would, on day one of practice, be operating an ambient AI scribe, an AI clinical decision support tool embedded in the electronic health record, an AI-augmented imaging interpretation system, and an AI triage augmentation tool in the waiting room. The institution had no credentialing framework for any of those tools. No competency requirement. No documented training. No peer review of AI-augmented clinical decisions. No privileging language in the medical staff bylaws. The committee approved the physician to use clinical AI by default, because the privileging structure had not caught up to the practice. That is true at most hospitals in the country in May of 2026, and it is the most important governance gap in clinical AI deployment that nobody is talking about.

The Joint Commission has now moved. The American Medical Association has now moved. The credentialing infrastructure has not. The hospitals that build the AI privileging framework in the next eighteen months will have a defensible governance posture. The hospitals that do not will be the case law.

What the Regulatory Landscape Now Requires

In June 2025, The Joint Commission announced a partnership with the Coalition for Health AI to co-develop AI playbooks, governance tools, and a Responsible Use of AI certification program. The first AI guidance from that partnership was released in fall 2025, and it landed exactly where credentialing committees need to look. The guidance places explicit obligations on healthcare organizations for AI governance, pre-deployment validation, post-deployment monitoring, bias assessment, voluntary blinded reporting of AI-related safety events, and — critically for credentialing — education and training of clinicians and staff on each deployed AI tool. The guidance is not yet binding, but it is the template for what the next accreditation cycle will inspect.

The American Medical Association moved in November 2025. The House of Delegates adopted formal policy supporting the development and dissemination of AI learning objectives and curricular toolkits aligned with existing AMA policy and Association of American Medical Colleges principles. The policy directs the AMA to collaborate with medical organizations to recognize AI literacy elements where appropriate, support continuing medical education to upskill the existing workforce, and advocate for funding and faculty-development resources for AI training. The AMA's framing is unambiguous: AI literacy is a competency. Competencies are credentialed.

The National Committee for Quality Assurance has tightened the credentialing structure overall. The 2025 NCQA Credentialing Product Suite shortened recredentialing windows, mandated monthly sanctions and license screening, and required primary source verification through direct API integrations rather than aggregator databases. The administrative bandwidth for new credentialing categories is being consumed by NCQA compliance, which is precisely the wrong moment for AI privileging to be a manual paper-based add-on. It has to be built natively into the same digital credentialing infrastructure or it will not happen.

That regulatory environment — Joint Commission AI guidance, AMA AI literacy policy, NCQA digital credentialing — is the new floor. The institutions that have not yet translated those documents into actionable medical staff bylaws are six to twelve months away from being out of step with their accrediting bodies. For the leadership framing on why an emergency physician is exactly the right person to drive that translation inside the institution, see my Medium essay Why Emergency Physicians Make the Best AI Governance Leaders.

What an AI Privileging Framework Actually Looks Like

Here is the framework I have drafted with the medical staff office at the institutional level. It is structured to slot into existing privileging packets without rewriting the entire bylaws.

A general AI literacy attestation is the foundation. Every credentialed physician at the institution attests, at initial credentialing and every two years at recredentialing, that they have completed a defined AI literacy module. The module is short — three to four hours of continuing medical education — and it covers the categories of clinical AI deployed in the institution, the failure modes those tools are known to exhibit, the override-and-edit responsibilities of the signing clinician, the documentation standard when AI output and clinician judgment diverge, the institutional reporting pathway for AI-related safety events, and the procurement-side governance the institution applies before any new tool is deployed. The attestation is a single line in the credentialing packet. The course satisfying it is institutionally sourced or accepted from a defined list of external providers. The AAMC and AMA curricular materials released through the 2025-2026 policy cycle will satisfy the external-provider pathway for most institutions.

Tool-specific privileging is layered on top of general literacy. For each high-stakes clinical AI tool deployed in the institution — ambient documentation scribes are the obvious first category, AI clinical decision support is the second, AI imaging interpretation is the third, AI triage and risk-stratification tools are the fourth — the institution defines a tool-specific competency standard. Minimum supervised cases, proctored encounters, documented competency assessment, and ongoing peer review at defined intervals. The structure is identical to the procedural privileging the institution already runs for central venous access or moderate sedation. The threshold is different and the assessment is different, but the framework is the same and the medical staff office already knows how to operate it.

Override documentation is the third element, and it is the one most institutions are missing entirely. The credentialing framework should require that any clinician operating a clinical AI tool whose output the clinician overrides on a specific patient encounter documents the override and the clinical reasoning in the chart. Not as a punitive measure. As a quality-data generator. The institution needs the override data to evaluate the AI tool against ground truth, and the medical staff bylaws are the right place to require its collection. The 2025 Joint Commission guidance explicitly endorses the development of post-deployment monitoring dashboards that depend on exactly this kind of override-and-divergence data.

Peer review of AI-augmented care is the fourth element. The existing peer review committee structure at most institutions reviews adverse events and outliers. It should also review, at a defined cadence, the cases in which clinical AI tools produced incorrect or borderline output and the clinician's response to that output. The review is not adversarial. It is the same quality-improvement function the committee already performs, applied to a new failure mode. The AI tool is part of the clinical environment. The clinician's response to the tool is part of the clinical practice. Both belong in peer review.

Continuing education is the fifth element. AI capabilities and AI failure modes evolve faster than most clinical domains. The credentialing framework should require a defined number of AI-focused continuing medical education hours per recredentialing cycle — eight to ten hours every two years is reasonable — with the credit categories drawn from the AMA and AAMC AI curricular framework as those materials mature through 2026 and 2027. The institution that requires its physicians to maintain ten hours of stroke continuing education per cycle has the same obligation to its AI literacy.

What the Medical Staff Office Needs to Do Now

Map every clinical AI tool currently deployed in the institution against the privileging framework above. Most institutions have between four and twelve clinical AI tools in active use, and the credentialing office can produce that inventory in a week if asked.

Update the medical staff bylaws to include the general AI literacy attestation. The bylaws revision is straightforward and does not require new committee structure. It requires a vote of the medical executive committee and a single line in the privileging packet.

Identify the two or three highest-stakes AI tools — ambient documentation scribes are nearly always one of them — and draft tool-specific privileging language for those tools first. Use the existing procedural privileging templates as the structural pattern. Define the minimum supervised cases, the competency assessment, and the peer review cadence. Submit to the credentialing committee. Iterate.

Stand up an AI literacy continuing medical education pathway. The simplest version is a quarterly grand rounds slot dedicated to AI in clinical practice, with attendance tracked and credit applied. The more developed version is an institutional micro-curriculum mapped to the AMA and AAMC framework, delivered through the existing learning management system, with credit auto-applied to the credentialing record.

Coordinate with the AI governance committee on override documentation standards. The two committees need to be reading the same data. The governance committee evaluates the tool; the credentialing committee evaluates the clinician's use of the tool; the peer review committee evaluates the divergent cases. The same documentation infrastructure feeds all three.

Dr. Chet's Take

I have served on hospital credentialing committees, military medical commissioning boards, and state-level medical practice review panels for twenty years. The pattern is consistent. New clinical capabilities get privileged on the same operational logic as old ones: define the competency, set the threshold, document the training, build in the peer review, recredential on a defined cadence. The framework is not novel. It is the most well-developed quality infrastructure in medicine. What is novel about clinical AI is that institutions are deploying the tools faster than the credentialing infrastructure is updating, and the gap is widening every quarter.

The credentialing committee that approves a new emergency physician in 2026 without any AI privileging language in the packet is treating clinical AI as if it were not part of the practice of medicine. It is. The Joint Commission has now said so. The AMA has now said so. The institutions that update their bylaws and their privileging templates in the next eighteen months will have a defensible posture when the accreditation cycle catches up. The institutions that do not will be operating without the governance scaffold the regulators are now writing. The medical staff office is the most important AI governance function nobody is talking about. It is past time to start.

AI Won't Wait. Neither Should You.

If your medical staff bylaws have no AI literacy attestation, no tool-specific competency framework, and no override-documentation requirement, your institution is operating clinical AI outside the credentialing structure that governs every other clinical capability you deliver. Consider enrolling in my course: AI in Emergency Medicine: Becoming AI Bulletproof. The course walks through the AI literacy attestation language we use in our institutional bylaws, the tool-specific privileging templates for ambient scribes and AI clinical decision support, the override documentation standard the institution should adopt, the peer review framework for AI-augmented care, and the continuing education pathway that satisfies the emerging AMA and AAMC AI curricular framework.

Learn more: AI in Emergency Medicine: Becoming AI Bulletproof

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.

Dr. 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.

Also on Medium: Read more from Dr. Shermer on Medium.

Sources

1. Katten / Joint Commission Guidance, "New Joint Commission Guidance On The Use Of Artificial Intelligence in Healthcare," October 2025, https://quickreads.ext.katten.com/post/102lqdt/new-joint-commission-guidance-on-the-use-of-artificial-intelligence-in-healthcare
2. American Hospital Association, "Joint Commission announces partnership to develop best practices for AI in health care," June 2025, https://www.aha.org/news/headline/2025-06-12-joint-commission-announces-partnership-develop-best-practices-ai-health-care
3. American Medical Association, "Boost health AI training across medical education continuum," November 2025, https://www.ama-assn.org/practice-management/digital-health/boost-health-ai-training-across-medical-education-continuum
4. MedCare MSO, "Medical Credentialing in 2026: What Has Changed and Why It Matters," April 2026, https://medcaremso.com/guide/how-to-navigate-medical-credentialing/
5. Medwave Billing & Credentialing, "Provider Credentialing in 2026: What's Changed and What Has Stayed the Same," March 2026, https://medwave.io/2026/03/credentialing-2026-updated-standards-best-practices-strategies/