The ACCME AI alert,
explained for academies.

Three documents in nine months

ACCME published three documents on AI in 2026.

January 30: guidance.

AI can support planning and drafting. Anything learners see still needs the provider’s review, valid content and independence from commercial influence. The guidance calls for human oversight of clinical outputs and careful handling of faculty material and learner data. Read the guidance.

April 14: the alert.

When AI interacts with learners in accredited CE, responsibility for what they see stays with the provider. ACCME lists seven controls to document and maintain. Inadequate oversight can threaten accreditation, including suspension. Read the alert.

September 22: content validity.

AI answering from a body of clinical content already validated by the provider can use the provider’s established review processes. Dynamic clinical answers generated from external sources may require independent assessment; ACCME decides when. The provider arranges and pays for an assessment, and material changes may require reassessment. Read the content-validity guidance.

Who this applies to

Accredited activities. If your academy awards credit through a joint provider, the joint provider holds the accreditation and carries the responsibility; expect them to ask what your AI does and how you validated it.

If your program doesn’t carry credit, these accreditation requirements do not apply to it. Checking an answer with your name on it still matters.

The seven requirements, in plain words

1. Keep product promotion out of the learning experience.
2. Validate the AI against clinical scenarios before launch and retain the results.
3. Define the clinical and educational limits of the tool.
4. Give named clinical reviewers authority to intervene or stop the activity.
5. Explain AI use and its limitations to learners.
6. Review outputs and interactions, correct errors, and repeat validation after system or prompt changes.
7. Require vendor agreements to give the provider the records and control needed for oversight.

These summarize the April alert. Keep the original with your activity documentation.

What to do this quarter, whichever vendor you use

Use this as a working checklist with your clinical lead. The suggested validation-set size is a starting point, not an ACCME-prescribed number.

• List every place an AI touches learners inside an accredited activity.
• Decide whether it answers from your validated content or dynamically generates answers from external sources. Discuss any assessment requirements with your provider.
• Write a scope statement in plain words.
• Build a validation set of thirty to fifty real questions learners ask, with the answer you expect and where it lives in your content.
• Name the clinician who owns it.
• Write the disclosure learners see.
• Put the interaction log and access to records in the contract.
• Decide what triggers a re-run of the validation set.

Seven questions to ask any vendor

1. Does it ever answer from outside our content? Under what conditions?
2. Can a reviewer click from any answer to the exact source?
3. Can we run our validation set before launch and export the results?
4. Can a named person on our side pause it?
5. Where is the disclosure shown, and can we edit it?
6. What is logged per question, and can we export all of it?
7. What happens to our validation when you change the model or the prompt?

What Bold does

Bold hosts your recordings and answers only from them, with the minute cited. Work through the seven controls with us and your clinical lead before launch, and agree the evidence your activity file needs.

For joint providers and consultants

This page is yours to send. The checklist and the vendor questions work with any product. If you want to work through the controls for your partners’ activity files, write to us.