On July 29, Hackensack Meridian Health became the first health system in the country to earn The Joint Commission’s Responsible Use of AI in Health Care certification. Most coverage treated it as a milestone for one New Jersey network. I read it as a deadline for everyone else, because of a single question that certification normalizes, and that most organizations cannot yet answer.
The question is: “Show me this model.”
Not your AI strategy deck. Not your innovation committee’s charter. One specific algorithm (the sepsis predictor, the imaging triage tool, the scheduling model) and its complete story: what it was approved to do, the performance number it was approved at, every check since, every incident, and the name of the human accountable for each decision.
I have spent my career in hospital operations, and I can tell you how that question lands in most organizations today. The evidence usually exists. The committee met; someone validated the model; IT watched the dashboards. What does not exist is the ability to produce that evidence, assembled and dated, inside a survey window. Decisions were made but not minuted. Monitoring happened but nobody wrote it down. The record lives in nine inboxes.
Real physician voices, twice a week
Free, and one click to unsubscribe.
Accreditation has always punished exactly this gap, in every domain it touches. AI is simply its newest domain.
There is a second, harder problem hiding inside the first. The evidence a certification asks for comes in two kinds. One kind can be drafted: policies, charters, disclosure standards (a focused week of work, whenever you choose to spend it). The other kind can only accumulate: quarterly performance checks against a locked baseline, committee minutes, training completions. Each of those has a date column, and date columns cannot be backdated. An organization that starts when the surveyor calls will have the drafted kind and lack the accumulated kind, and the accumulated kind is what proves governance actually operated.
That is why the first certification matters beyond New Jersey. Voluntary certifications in health care have a familiar life cycle: first mover, then fast followers, then the board meeting where someone slides the press release across the table and asks why a peer got there first. From that meeting forward, the organizations that look prepared will not be the ones that worked hardest in the final quarter. They will be the ones whose clocks started earliest.
One more uncomfortable truth, about what monitoring must now mean. The metric most organizations track (overall accuracy, usually AUROC) can hold perfectly steady while a model quietly fails one subgroup of patients. Case mix shifts, a documentation template changes, and sensitivity for one language-preference group slides while the average stays green. A monitoring program that watches only the headline number will call that model healthy. Equity monitoring against locally chosen thresholds is not an advanced feature of AI governance; it is the part that protects the patients averages hide.
None of this requires waiting for the standards manual, a consultant, or a budget cycle. The preparation order is unglamorous: Inventory what you actually run, including the AI that arrived inside vendor upgrades rather than purchase orders. Charter the committee, because minutes only accumulate forward. Lock a baseline on your highest-risk model and check it quarterly, in writing, subgroups included. Train people and record it. Draft the documents last, because documents are the only part that can be rushed.
The quiet period between a certification’s announcement and its arrival in your board’s agenda is the only time this preparation is cheap. On July 29, that period started ending.
Neel Chauhan is a physician-executive and the founder of The Healthcare AI Institute, an independent publisher and software maker for health care AI governance.
He is the author of fourteen operational titles on health care AI, covering governance, ambient documentation, revenue cycle, scheduling, and primary care operations. He also maintains the Institute’s open resources: the Transparency Registry of FDA AI clearances, the Agent Privileging Framework for governing AI that acts, and the working reference on The Joint Commission’s Responsible Use of AI in Healthcare certification.
The Institute accepts no vendor sponsorship, holds no vendor equity, and takes no referral fees. He shares updates on LinkedIn.
