Generative AI is already in clinical hands, sanctioned or not. Wolters Kluwer Health’s 2026 Future Ready Healthcare survey found 74% of physicians and 70% of nurses use generative AI at least weekly — and that 77% of physicians see hallucinations as a significant risk while 92% want a human expert validating AI-generated clinical content. That makes healthcare AI literacy a governance and workforce to workflow question, not just a training one.
In a featured session sponsored by Wolters Kluwer at Becker’s 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health on Sept. 15, physician and informatics leaders examined what literacy requires at every level. The panelists were:
- Brian Hasselfeld, MD, executive medical director of digital health and innovation at Johns Hopkins Medicine (Baltimore)
- John-Paul Mead, MD, CMIO at Centralus Health (Ithaca, N.Y.)
- Joseph Thomas, MD, associate CMIO at Baptist Memorial Health Care (Memphis, Tenn.)
- Amanda Heidemann, MD, senior clinical content consultant for clinical effectiveness at Wolters Kluwer Health
Below are four takeaways from the discussion.
Note: Quotes have been edited for length and clarity.
1. Literacy is role-specific and challenging to define.
Dr. Hasselfeld said the term “AI literacy” presumes AI is generalizable when it is not. It is as much labor as technology, he said, so literacy blends technology skills with the management skills of delegating, supervising and knowing when not to use a tool.
Building on that, Dr. Thomas noted competence looks different at every level; a nurse needs to know a sepsis predictive model is a guide, not a diagnosis. Executives face a different test. “It’s just not a shiny new tool anymore,” Dr. Thomas said. “It is asking the question about what is this data being trained on and, most importantly, will this model work for our population?”
2. Bring the AI-native workforce out of the shadows, and hear the skeptics.
Dr. Hasselfeld said Johns Hopkins invested in internal platforms and shared team channels so self-taught employees can show what they built over the past week with AI, get recognized for it and have the skill reused across teams rather than hidden.
Dr. Heidemann offered a counterpoint: it’s important to identify the skeptics too. “Are they informal thought leaders that have the potential to make or break a project?” she said. In her work at Wolters Kluwer, she’s observed that one-on-one conversations with skeptics have helped to surface project risks not previously considered.
3. Governance must be fluid.
Last month, Dr. Mead briefed his board’s audit and quality-and-safety committees, who asked how AI risk is being controlled and whether clinicians know when they’re using it. He said the board was reassured that the system already sorts tools by risk — fully autonomous applications are highest risk and personal productivity tools lowest.
Clinical tools that appear to give answers fall in between: they hold the most promise but need the closest scrutiny, Dr. Mead said. The board also asked that every tool link to its evidence, since clinicians “need to be skeptical and double check and triple check these things if you’re making medical decisions,” he added.
The discussion repeatedly returned to transparency and traceability. Panelists noted that clinicians and governing bodies increasingly want to understand not only what an AI system recommends, but also the evidence, reasoning boundaries and oversight mechanisms behind those recommendations.
Citing a colleague’s published framing, Dr. Hasselfeld said the field is asking for “an autopilot without knowing how to certify the pilot.” With no regulatory paradigm, he said, risk assessment falls to institutions, and the risk paradigm Johns Hopkins uses today “was different six months ago, and it’s going to be different six months from now.”
4. Evaluate the guardrails — not just the evidence.
When it comes to trusting a tool’s output, Dr. Mead said the starting point is the data; internal data must be validated before a model is layered on top, or AI will amplify existing errors. Dr. Thomas agreed, adding that local validation and a vendor’s willingness to explain what’s under the hood are what build trust.
Dr. Heidemann emphasized that guardrails, the instructions that tell a model when to answer and when not to, matter as much as the evidence beneath it. “If they won’t talk to you about how their guardrails are set, or who is deciding on the guardrails, that’s something that’s really important to consider,” she said.
Panelists suggested that trust in clinical AI depends on more than access to evidence alone. Health systems increasingly need visibility into how expert guidance, validation processes and operational guardrails govern the way AI-generated responses are produced and communicated.
When asked by an audience member what AI tools should be measured against, Dr. Hasselfeld said no one has settled the question: gold-standard care, the median care most patients receive, or no care at all.
One action for the next 12 months
The panel closed with one step organizations can take in the next year to strengthen AI literacy and innovation, and Dr. Thomas and Dr. Mead gave the same answer: inventory the AI already running in the organization.
“We sometimes don’t even know that we have AI somewhere,” Dr. Thomas said. From there, Dr. Mead suggested picking simple use cases that make staff’s lives easier, like sparing clinicians late-night work from home, and using them as the model for the next five rollouts. Dr. Hasselfeld split his answer across strategy, oversight and procurement. He said managers should set AI-anchored goals, oversight should keep building out risk assessment and post-deployment monitoring, and procurement should put the infrastructure behind both.
Dr. Heidemann offered a patient-focused, point-of-care test for both current and prospective tools: trace each one’s line of sight to the patient. “There are quite a few AI tools that are out there that actually make things worse for patients,” she said, citing AI-driven prior authorization, where “the bots are fighting each other about whether something’s going to be covered.”
“I think it’s just important to understand where the patient fits in that decision, and I would encourage people to look for that,” she said.
As health systems expand AI adoption, leaders will need to evaluate whether tools are grounded in trusted evidence, informed by expert judgment, transparent about their sources and accountable for their performance. Ultimately, the goal is not simply broader AI adoption, but clinical intelligence that remains reliable, explainable and evidence-based, enabling better decisions and better care for patients.
Becker’s 3rd Annual Fall Chief Pharmacy Officer Summit is the top tier event that the people leading the future of pharmacy convene. This 3-day event is built specifically for Chief Pharmacy Officers and senior pharmacy leaders. The sessions examine the workforce, financial and clinical priorities shaping pharmacy today. Becker’s is where candid discussions flourish and new ideas thrive. Becker's 3rd Annual Fall Chief Pharmacy Officer Summit
