Spend :01 of your time each Monday morning as Twelve:01 delivers timely tools, trends, strategies, and/or compliance insights for the CME/CE enterprise.
AI tools are increasingly able to connect directly to authoritative sources, including PubMed, DailyMed, ClinicalTrials.gov, and CMS data, making it easier to locate and synthesize healthcare information. However, access to a trusted source does not guarantee that the model selected the right evidence, interpreted it correctly, or preserved important context. In accredited CME/CE, treat source-connected outputs as research support, not validated educational content. Qualified reviewers should confirm citations, compare claims with the original materials, assess balance and bias, and document the review process. ACCME guidance reinforces that AI-assisted materials remain subject to the same expectations for accuracy, independence, and human oversight as content developed by human authors.
ACCME’s Joint Providership Policy allows an accredited provider and one or more nonaccredited organizations (the joint provider) to plan, implement, and evaluate accredited CME/CE together, but it does not divide accreditation accountability. The accredited provider remains responsible for content validity, independence, financial relationship collection, review, identification and mitigation, required statements, documentation, and compliance with criteria, standards and policies. That responsibility continues even when the joint provider manages faculty, marketing, registration and/or technology. The accredited provider must demonstrate compliance through written documentation and inform learners of the joint providership relationship through the appropriate accreditation statement. Before entering into joint providership, the accredited provider should also verify the joint provider’s eligibility.
A new Harvard Data Science Review article argues that AI safety cannot be established through model performance, benchmarks, guardrails, or other technical controls alone; it must be evaluated across the full sociotechnical system in which AI operates. Drawing lessons from failures in aviation, nuclear power, healthcare, and other complex environments, the authors highlight organizational risks such as rushed implementation, fragmented accountability, weak risk communication, and “symbolic compliance” that does not meaningfully reduce risk. The article offers an interesting assessment and a useful framework for CME/CE professionals considering AI governance: evaluating a tool should include not only what the technology can do, but who uses it, where human review occurs, how decisions are documented, and who owns responsibility when something goes wrong. As organizations move AI into needs assessment, content development, faculty support, and other workflows, governance should focus on the safety of the entire workflow, not just the model’s reliability.
