Healthcare AI Governance Frameworks

Your patients are already using AI. Do you know how?

Governance briefings for hospital leaders, from a patient who looks into medicine, not up at it. AI that frees the patient frees the clinician.

Compliant by Default Was Always a Sales Claim

Every query embedded in your EHR used to carry an unspoken promise: if the vendor built it, it was compliant. Brian Fengler — an emergency physician and co-founder of EvidenceCare — reports that promise just became unenforceable.

In August, AHIMA and ACDIS released the 2026 update to their Guidelines for Achieving a Compliant Query Practice, shaped by feedback from nearly 200 individuals and organizations. The operating principle: whether a query comes from a person or from software, the compliance standard doesn't change. The sentence that should end a decade of sales decks: organizations should not assume a vendor-supplied or EHR-integrated tool produces compliant queries by default.

A query is what it does, not what it's called

The guidelines define a query by function. Any communication that presents a provider with a specific diagnosis or documentation option for a specific patient encounter is a query — regardless of how the platform labels it. A prompt, a nudge, an advisory, an alert. Calling it something softer doesn't exempt it. "Here are the sepsis criteria" is reference material. "Consider documenting severe sepsis for this encounter" is a query, full stop, whatever UI it's wrapped in.

That line matters because it closes the gap AI vendors have been quietly living in. Templates can't auto-populate the answer field. Formatting can't steer a clinician toward a preferred response. And if an automated query doesn't produce the diagnosis an organization wanted, sending a manual follow-up solely to override the provider's answer is noncompliant — unless new clinical evidence actually exists.

Accountability doesn't transfer with the task

Here's the sentence hospital leadership needs taped above every AI procurement meeting: organizational process owners retain accountability for every query delivered to a provider, including those generated through automated or AI-assisted processes. The burden of proof sits with the health system that bought the tool, not the company that sold it.

Regulators are already there. The HHS Office of Inspector General's March 2026 Medicare Advantage guidance names "AI-generated prompts encouraging unsupported coding" as a potentially abusive risk adjustment practice. And query practice drew real enforcement before AI entered the picture at all — Independent Health's 2024 settlement, up to $98 million, over allegations that a subsidiary queried physicians to support higher risk scores. No AI involved. The pattern regulators are watching for already existed. AI didn't invent the risk. It scaled it, then packaged it as a feature.

Clinician authority is exactly what's at stake here, even though nobody's using that language. A query that pre-selects an answer, that can't be traced back to real clinical indicators, that measures success by how rarely it escalates to a human — that's not assistance. That's a system quietly making the call while leaving the clinician's name on the record. The 2026 guidelines are, underneath the compliance language, a demand that the decision stay visibly and provably human.

Fengler's four questions belong in every AI-enabled documentation evaluation from here forward: How does the tool prevent a leading query, concretely? Which tools in your environment — EHR alerts included — actually meet the legal definition of a query? Who reviews before delivery, or how often do you audit after? What's your process for correcting one once you find it?

"Compliant by default" was never a technical property. It was a marketing claim. Now it's a claim the adopting organization has to prove, one audit cycle at a time.


Source: Brian Fengler, MD, "Is Your AI Query Tool Compliant by Default?," The Health Care Blog, October 14, 2026.