Welcome Back!
I took a short hiatus to plan my next move. While I was away, AI continued consolidating. Twelve major health systems joined forces with Aidoc to co-design the next generation of diagnostic AI. Anthropic is pushing hard into clinical markets ahead of its IPO. Epic is quietly rolling out AI tools that are already live in emergency departments. And a new analysis puts AI note hallucination rates at 31%.
This week, I’m debuting my “AI skill of the week.” AI is getting easier and easier to access, and its important to stay ahead. Clinicians NEED to learn how to use AI. Each week, I will take a deep dive into what clinicians should know to understand AI, and start building for themselves.
The first skill I’m teaching is how to create AI-generated patient education materials. Every week, we tell patients the same things. We repeat ourselves over and over again. But how much is the patient retaining? Most of the time, they need to be reminded again. Pamphlets, flyers, booklets, and other types of patient education materials allow patients to re-read what we have told them at home. With AI, its easier than ever to create them at home.
Stay tuned to the end and I’ll show you how easy it is.
📰 Latest News
12 health systems just formed a diagnostic AI consortium covering 20 million patients: Advocate Health, Cedars-Sinai, Mount Sinai, Northwestern Medicine, Northwell Health, and seven other major systems announced a formal Diagnostic AI Consortium with Aidoc to co-design AI workflows, share real-time feedback, and build evidence under IRB oversight.
Anthropic is making a major healthcare push before its IPO: Ahead of a planned September or October IPO, Anthropic is positioning Claude for Healthcare, Claude for Life Sciences, and Claude Science as core enterprise offerings. For clinical AI buyers, that means Anthropic is now a serious enterprise vendor, not just a research lab.
Epic's ambient AI charting tool is live in emergency departments: FMOL Health became the first Epic client to deploy ambient AI charting in EDs, with the tool listening to patient encounters and drafting clinical notes for physician review. Tampa General is simultaneously piloting Ask Art, a conversational AI that lets clinicians query a patient's labs, medications, and clinical history in plain language.
🚀 Technical Advancements
From NVIDIA
NVIDIA's healthcare AI survey finds the field has moved from experimentation to ROI: NVIDIA's 2026 State of AI in Healthcare survey shows that health systems are no longer asking whether AI works, they are measuring what it returns. Medical imaging and drug discovery lead on demonstrated ROI, with the industry shifting from pilot programs to production deployment at scale.
AI drug discovery is entering pivotal clinical trials - no longer speculative: Multiple AI-designed drug candidates are now in pivotal trials with clinical readouts expected throughout 2026. A landmark Nvidia and Eli Lilly co-innovation lab is building a continuous learning system that connects agentic wet labs with computational dry labs around the clock, compressing timelines that once took years into months.
Ambient AI documentation is expanding beyond physicians to nursing workflows: Epic, in partnership with Microsoft and Abridge, is piloting ambient AI that documents nurse-patient interactions and auto-populates nursing assessments directly into EHR flowsheets. The expansion matters because nurses generate a significant share of clinical documentation, and their workflows have historically been underserved by AI tools designed primarily for physician use.
🔬 Research Spotlight
AI clinical notes contain hallucinated content in 31% of drafts, higher than physician error rates: A 2026 benchmark analysis found that AI-generated ambient notes contained hallucinated clinical content in nearly one in three drafts, compared to a 20% error rate in physician-authored notes. If an incorrect fact enters the input, AI models propagate it in up to 83% of downstream outputs, making input quality and clinician review non-negotiable parts of any AI documentation workflow.
Physician AI adoption has doubled since 2023, but so has the variation in how it is used: More than 80% of physicians now report using AI in professional practice, up from 40% in 2023, and more than three-quarters say it improves their ability to care for patients. The data also surfaces a significant gap: adoption is high but structured workflows and documentation of AI use remain rare, meaning most AI use is still happening without accountability frameworks.
The Mayo Clinic Platform is showing how to validate clinical AI at scale, before it harms anyone: A February 2026 paper in npj Health Systems describes how Mayo's secure, multi-institutional, de-identified clinical data platform enables efficient cohort identification, AI model development, and real-world evidence generation across institutions. The platform represents a practical infrastructure model for health systems that want to develop and validate AI internally rather than buying vendor-only solutions.
⚖ Ethics and Policy
California's AI Transparency Act took effect August 2, and it is just the beginning of a complicated regulatory era: California's SB 942 now requires developers of generative AI systems with over one million monthly users to provide tools that let users identify AI-generated content. For healthcare organizations using these platforms in clinical workflows, the law raises practical questions about disclosure obligations, patient communication, and documentation standards, questions that most health systems do not yet have clear policies to answer, and that will only multiply as more states follow California's lead.
🧠 AI Skill of the Week
Bad Example of Patient Education
Creating Patient Education Materials with AI
The Skill: Use AI to turn a complex medication, diagnosis, or procedure into a clear, plain-language handout your patient can actually understand and take home.
Why It Matters: Health literacy is one of the most underaddressed barriers in medicine: only 12% of U.S. adults have proficient health literacy, yet most clinical materials are written at a 10th grade reading level or higher. As AI tools expand across documentation and diagnostics, they also create a practical opportunity to dramatically improve patient communication. A well-crafted AI-generated handout can close that gap in minutes, not hours.
Try It Now:
"I need a one-page patient education handout about [medication name or procedure]. My patient is [age], and I want this written at a 6th grade reading level with no medical jargon. Include: (1) what this medication or procedure is in simple terms, (2) why their doctor recommended it, (3) what to expect, (4) important side effects or warning signs to watch for, and (5) when to call the office. Use short sentences and plain language. Format it so it is easy to read and could be printed as a handout."
💭 Final Thoughts
The Aidoc consortium and Epic's emergency department rollout are the clearest signals yet that clinical AI is moving from institution-by-institution experiments to coordinated, large-scale deployment. Twelve major health systems designing AI workflows together under IRB oversight is a materially different thing than twelve health systems each running their own pilots. The results, when they come, will carry weight that single-institution studies cannot.
At the same time, the hallucination data is a useful counterweight to the momentum. AI note drafts contain errors at a higher rate than physician-authored notes. That does not mean AI documentation tools are not worth using, the time savings and burnout reduction data are real. It means the review step is not optional. Every workflow that deploys AI for clinical documentation needs a built-in, non-negotiable review process before anything enters the permanent record.
That is what this week's skill addresses directly. A 31% hallucination rate is not a reason to avoid AI documentation tools. It is a reason to use them with a checklist.
Regards,
Chris Massey, MD
"The art of medicine was to be properly learned only from its practice and its exercise."
What AI skills would you like to learn?
Let me know what you’d like to learn and I’ll include it in a future newsletter.
email me @ [email protected]
Disclaimer: This newsletter is for educational and informational purposes only and does not constitute medical advice. Readers should review primary sources and follow applicable clinical guidelines and institutional policies before implementing any changes. Always de-identify patient data and review all outputs for accuracy.