Fellows Already Use AI Daily, Yet Formal Training Remains Rare

Key Takeaways

• A multi-institutional hematology/oncology fellow survey presented at ASCO26 put numbers on a familiar AI problem: use is already common, but formal training is rare.
• For CME providers, the near-term opportunity is not another broad AI overview. It is role-specific training in appraisal, safe use, and documented competency.
• The signal is oncology-led and fellow-specific, but the design question travels to other data-heavy training environments.

A multi-institutional ASCO26 survey put a number on the AI training gap: many hematology/oncology fellows are already using LLMs for education, while only a small minority report formal AI training. This is a narrow fellow-level signal, but it gives CME teams something more concrete than generalized concern about AI adoption.

The gap is no longer abstract

The clearest signal came from ASCO26 posts around a multi-center survey of AI use among hematology/oncology fellows. One clinician post summarized the finding plainly: fellows are using AI for learning, but “only 8%” are getting any training (source). A second clinician post highlighted how AI tools now sit beside conventional learning resources: “The survey highlights popular resources: NCCN guidelines (92%), question banks (86%), reference websites (86%), and notably, AI tools like ChatGPT (74%).” (source)

The companion context matters. A society podcast previewing ASCO26 placed the survey inside a broader medical education track, naming “AI and Hemoc Fellowship Training, a multi-center survey of education attitudes and clinical use” as an oral presentation (source). That does not make this a cross-specialty consensus. It is hematology/oncology fellow data, surfaced during a major oncology meeting. But the provider implication is broader: when trainees already use AI to clarify concepts, summarize literature, and learn emerging research, CME cannot treat AI literacy as a future-facing elective.

This extends the provider problem we flagged in an earlier brief on AI governance training lagging real-world tool use. What changed this week is the specificity. The gap is not just “clinicians are using AI without enough supervision.” It is a measurable mismatch between adoption, confidence, critical appraisal, and formal instruction inside a defined training population.

For CME providers, that shifts the curriculum question. A generic module on “AI in medicine” is unlikely to be enough. Fellows need to show they can judge AI-generated summaries, recognize when a tool is smoothing over uncertainty, understand where outputs fit in the evidence hierarchy, and decide when AI use is inappropriate for a clinical or educational task. The concrete implication: build AI education around observable appraisal behaviors, not just awareness of tools.

What CME Teams Should Reconsider

• Audit existing AI education for whether it teaches critical appraisal of AI outputs, not just risks, definitions, or tool demonstrations.
• Separate learner objectives by role: fellows using AI for study and literature review need different practice tasks than faculty using AI for supervision or documentation.
• If partnering with fellowship programs, define what evidence of competency will look like before choosing format or credit structure.

The design question

The important question is not whether fellows will use AI. In this survey signal, they already do. The question for CME teams is whether formal education can catch up with the actual use case: trainees applying AI inside learning, literature review, and early clinical reasoning before programs have agreed on what competent use looks like.