Communication failures contribute to more than 60% of hospital-based adverse events. That's not a new number -  it's been documented for years. What's new is that AI has created the political conditions to finally do something about it.

In February 2026, ECRI named the misuse of general-purpose AI chatbots as healthcare's number one technology hazard. The finding was specific: AI-generated output that sounds authoritative causes clinicians and staff to act without verification, because no review layer exists to catch errors before they reach a patient. Health system leaders are now building AI governance frameworks. But most of those frameworks are addressing the AI layer while leaving a pre-existing layer completely untouched: clinical communication itself.

When a clinician communicates a diagnosis ambiguously, fails to confirm what a patient actually understood, or conducts a handover that creates downstream confusion, the interaction happens and then the evidence of it disappears. Roughly 67% of communication errors occur at handoffs - the moments of highest transfer risk. None of it is systematically reviewed. No signal reaches the quality team. No feedback loop closes.

This isn't a training problem. It's an infrastructure problem. Workforce training programs build communication skills in controlled settings. What most health systems lack is any mechanism for observing and improving communication in real clinical practice, at scale, over time. An organization can score well on simulation assessments and still have undetected communication failures occurring in its wards every day.

Most AI governance efforts are focused on the output layer: review before acting on AI-generated clinical notes, documentation, or treatment suggestions. That's the right instinct. But it addresses the symptom, not the structural condition.

The structural condition is that healthcare organizations have never built review infrastructure for clinical communication, whether human or AI, in real-world practice. Only 16% of health systems currently have an enterprise-wide AI governance strategy. Those that are building one are layering it on top of a system that already didn't review its own communication.

The most sophisticated AI review framework won't catch communication failures that occur before AI is even involved. And it won't build the organizational muscle: the workflows, the feedback culture, and the data infrastructure needed to review AI output reliably once it does arrive.

Health systems that have made progress here share a pattern. They didn't attempt enterprise-wide change. They identified one workflow (one department, one care transition, one cohort) and built a structured review process around it first. Consent frameworks, recording infrastructure, and feedback processes were developed incrementally, then expanded.

The evidence on what structured communication review produces is consistent: observation of clinical communication, followed by repeated and structured feedback, generates measurable improvement in behaviour. The mechanism isn't complicated. You cannot improve what you cannot see. Communication that is observed and reviewed develops differently than communication that disappears the moment it occurs.

For years, the case for investing in clinical communication review struggled to travel above a departmental level. Quality improvement teams understood it. Executives were harder to move.

That's changed. A credentialed patient safety body has named unreviewed communication as a top-tier clinical hazard, and AI has made the problem visible to people who were not previously looking at it. Risk management conversations find budget in ways that training conversations don't.

Health systems building AI governance frameworks in 2026 have a window to also address the pre-existing gap. The infrastructure is largely the same: consent and recording workflows, structured review processes, and feedback systems that generate longitudinal data on communication quality. Organisations that build this for clinical communication find they have also built the foundation for responsible AI oversight in clinical settings, because the capability they're developing is review at scale, not just review of AI.

The organisations ahead of this problem didn't wait for AI to make the case. They understood, before the conversation started, that clinical communication is clinical behaviour, and clinical behaviour needs to be observed, reviewed, and improved over time. That foundation is now an AI governance advantage.