AAMC Is Building National AI Competencies for Medical Education. The Report Lands This Fall.

The AAMC is developing a formal set of AI competencies spanning the entire medical learning continuum, from pre-clerkship students through practicing physicians, in partnership with international medical education bodies including AMEE and IAMSE. A volunteer advisory committee, selected through an open call for self-nominations, is reviewing draft competencies now. The target for a final report: fall 2026.

That timeline matters more than it might first appear. For programs that have spent the last two years improvising AI policy case by case, largely without a shared standard to build against, this is the closest thing to a finish line that's been publicly dated so far.

Why This Gap Has Been Getting Harder to Ignore

Physician AI adoption has moved faster than the governance built to guide it. Recent AMA survey data cited at the American Thoracic Society's 2026 annual meeting found that 81% of U.S. physicians now use AI tools in clinical care, more than double the rate from just a few years ago, with the most common uses including summarizing research, drafting discharge instructions and progress notes, and generating chart summaries.

That adoption curve has outpaced the research needed to understand what it's doing to trainees. Research from the NOHARM network found that 76.6% of severe AI errors in clinical settings were errors of omission, not incorrect or dangerous suggestions, meaning the failure mode that actually matters most is the answer AI never gave, not the wrong one it did. That's a hard thing to teach residents to watch for without a shared framework describing what "watching for it" is supposed to look like.

Right now, "should residents use AI to generate a differential" is a question competency committees are fighting about in real time, not one with a standard answer. One recent account described a committee meeting where faculty were giving residents feedback that they should be using AI this way, only for other committee members to interject: wait, do we actually want them doing that? A national framework won't end that kind of debate outright, but it will give program directors something to calibrate against instead of building guardrails from scratch, program by program, with no shared reference point.

Building on Ground Already Broken

The AAMC's AI-specific effort extends a model it has used before. In December 2024, AAMC, AACOM, and ACGME co-released Foundational Competencies for undergraduate medical education, covering six core domains: professionalism, patient care and procedural skills, medical knowledge, practice-based learning and improvement, interpersonal and communication skills, and systems-based practice. Those competencies gave schools a shared outcome set regardless of degree type or specialty. The AI-specific competencies now in draft extend that same architecture into a domain none of the original six anticipated.

At a Penn LDI panel this past academic year, faculty from Stanford, Northwestern, and NYU pushed the ask further, arguing that AAMC, ACGME, and LCME need to align on shared AI competencies specifically across the UME-to-GME transition, not just within each body's own lane. That's not yet a confirmed joint initiative bridging all three bodies. It's a recommendation from educators on the ground, and it's worth distinguishing from AAMC's own dated effort. But it signals where the pressure for alignment is coming from, and it's unlikely to go away quietly.

A Framework Already Filling the Gap

One framework already getting cited as a stopgap while the national standard is drafted: DEFT-AI, built around five components: diagnosis, evidence, feedback, teaching, and AI engagement. It was introduced in a New England Journal of Medicine article and flagged at the Penn panel by NYU's Verity Schaye as a practical tool for preventing overreliance on AI while it's still being used to support judgment rather than replace it. Frameworks like this exist precisely because programs can't wait until fall 2026 to decide how residents should be taught to use these tools; they need something to hand faculty now.

The Open Question Is Timing

A fall 2026 target report means programs are heading into another full academic year without a national standard, which means the governance gap doesn't close on its own, it just gets a firmer end date. Programs building internal AI oversight now aren't waiting to get ahead of the curve. They're the ones the eventual standard will likely end up resembling, because they're the ones already doing the work the AAMC's advisory committee is still drafting guidance for.

Worth watching: whether the AAMC's fall report explicitly addresses the UME-GME transition point the Penn panelists flagged, or leaves that cross-body alignment for a separate effort.

This is the exact gap edYOU was built to close. While national bodies work through what AI governance in medical education should look like, edYOU's Provost platform already treats hallucination governance and clinical judgment preservation as architectural first principles, not policy added after the fact. If your program is building internal AI oversight now instead of waiting for fall, see how Provost supports that work.