The AAMC Named Six AI Competencies. Teaching Them Is Still Your Problem.
Six days before the AAMC published its AI competency framework, another document emerged with less fanfare and more impact.
Barr and colleagues surveyed 15,259 family medicine residents. The study was published in the journal Family Medicine on 21 October 2026, with a 99.6% response rate. Seventy-one percent reported using an AI tool in the previous month. Only 36.3% said they had received any education on AI in the previous six months.
That’s the environment the framework landed in. Residents are already using the tools. Most of them taught themselves.
What the AAMC actually published
On 27 October 2026, the AAMC issued AI Competencies Across the Learning Continuum. It defines six domains: AI Knowledge for Practice. AI-Assisted Patient Care and Clinical Reasoning. Communicating About AI and Interprofessional Collaboration. Professionalism and Ethics. Practice-Based AI Learning and Improvement. Systems-Based AI Practice.
The framework addresses educators, learners, institutions, and health systems. It names them directly and says the competencies are designed to help those groups incorporate AI into medical education and clinical practice. We wrote about this framework while it was still in development; this is what actually arrived.
What the AAMC doesn’t do is tell you how. It describes its own work as “an educational framework, an invitation, and a historical marker.” It states that certain system-level competencies may be aspirational. It augments existing competency frameworks rather than replacing them. It isn’t an accreditation requirement and doesn’t claim to be one.
That’s the right call for a framework. A national body names the destination and leaves the route to individual programs. It also means the hard part now sits on your desk.
The capacity problem nobody has solved
A competency framework presupposes that someone can teach the competencies. For residency programs already balancing clinical training, board preparation, remediation, and faculty responsibilities, that raises a practical question: who has the time and expertise to teach AI use, assess it, and oversee it?
The AAMC’s revision notes say the process strengthened its focus on faculty development. They don’t provide a faculty development plan, because that falls outside what a framework does. So each program has to work out which faculty members are qualified to judge whether a resident’s use of a clinical AI tool was appropriate.
For most programs in November 2026, the honest answer is that faculty learned these tools the same way the residents did. The 36.3% figure reflects both what residents have been taught and what their instructors could offer.
This pattern runs throughout graduate medical education right now. We looked at it from the governance side in our article on the AI scribe guidance gap: the bodies that set standards have moved more slowly than the tools. The teaching gap is the same gap, one layer down.
Three things your program can do without new national guidance
Anchor the tools to your own curriculum
A general-purpose assistant trained on the open internet answers a resident’s question with whatever it has. A system grounded in your curriculum, your protocols, and your institution’s materials answers from the body of knowledge you are accountable for.
The difference shows up when the model is confidently wrong, which is also when a resident is least equipped to catch it.
We developed this argument further in why generic AI platforms fail in medical education.
Establish academic integrity rules before you need them
Most programs are still working from policies written for a different problem. Decide now what appropriate use looks like in your program, in writing, with specific examples.
A rule nobody saw until the disciplinary hearing will not survive the hearing.
Build faculty oversight in as a design requirement
If a system can’t show a supervisor what a learner asked, what the tool answered, and where that answer came from, it can’t be supervised. Oversight isn’t something you add later.
It determines whether the tool can be part of a training program at all.
All three cost money, which most programs discover last. We set out what that looks like as a line item in the case for a dedicated GME AI budget.
Next steps
Three questions worth putting to your program leadership before the end of this quarter:
Which of the six AAMC domains do we already assess, even informally, and which have we never once looked at?
If a resident used an AI tool badly tomorrow, who on our faculty would be qualified to say so, and what would they point to?
What is our written rule on appropriate use, and when did a resident last read it?
The framework was published on 27 October. Residents were already using these tools before it existed. The gap between 71.1% and 36.3% in the family medicine survey is the problem worth examining, and no framework closes it for you.
The next challenge isn't defining AI competencies. It's finding a practical way to teach them, reinforce them, and give faculty enough visibility to know whether residents are actually learning.
At edYOU, we've built Academic Intelligence around these principles. Our curriculum-grounded learning systems combine structured educational content, personalized learning support, and faculty-facing insights to help institutions bring AI into medical education without removing educators from the center of the process. The objective isn't to replace faculty judgment or add another responsibility to an already demanding workload. It's to give programs a more manageable way to support resident learning while maintaining academic oversight.