Our Take
Clinicians are solving real problems with unsanctioned tools; administrators are burned out on pilots. The gap between demand and supply is where breach risk lives.
Why it matters
Over 80% of physicians now use AI in clinical work (per AMA), but most health systems lack governance frameworks or integrated solutions to support it safely. This misalignment between user demand and institutional readiness is the immediate challenge for digital leaders, not vendor evaluation fatigue.
Do this week
CIO/Chief Medical Officer: Audit which AI tools clinicians are actively using outside your sanctioned stack this month, then fast-track enterprise solutions for the top three use cases (likely documentation, chart summarization, inbox triage) before year-end so you can close governance gaps.
Clinicians are ahead of their institutions
According to the American Medical Association's Physician AI Sentiment Report, more than 80% of physicians now report using AI in some professional capacity (per AMA). This represents a dramatic increase from prior years. The adoption spans documentation support, clinical research, coding assistance, translation, and decision support.
Physicians report clear advantages for patient care. A strong majority believe AI can help reduce administrative burden, one of the primary drivers of burnout. The same AMA report found that 73% of physicians see growing opportunities for AI to automate administrative tasks.
Yet among healthcare executives and digital transformation leaders, sentiment is different. Nine out of ten decision-makers report generative AI pilot fatigue, driven by vendor proliferation, constant "AI-powered" pitches, and lack of evaluation tools.
The result: clinicians are not resisting AI. They're requesting it to eliminate friction from existing workflows. But many are using unsupported, unsanctioned tools because enterprise alternatives don't yet exist or don't fit their daily work.
The governance gap is where risk and opportunity live
When clinicians choose unsanctioned AI tools to solve real problems, several risks emerge in parallel. Data may leave secure environments. Data models may lack validation workflows. Outputs may not integrate back into the EHR in structured form. Standards become inconsistent. These are not abstract compliance concerns—they are immediate breach vectors.
But the opportunity is sharper: health systems that move fast to provide "invisible AI" (tools that work inside existing workflows, not in new tabs or logins) will both reduce clinician burnout and contain the shadow-IT problem. Physicians are optimistic about AI when it eases administrative tasks, reduces documentation time, and improves efficiency without adding friction.
The first wave of scaled enterprise AI should focus on patient medical history aggregation, automated chart summarization, structured synthesis of external medical records, ambient documentation integrated into the EHR, inbox triage and message routing, and coding and prior authorization support embedded within workflows.
How to close the gap without chasing headlines
Health systems do not need to chase every AI headline or move recklessly. But they do need to realize that multi-year roadmaps will leave them behind competitors whose clinicians have already solved the problem with unsupported tools.
Start with what clinicians are already trying to solve. Focus early efforts on reducing documentation burden, inbox overload, and manual chart review. These are the areas where physicians already see value and where AMA data shows the strongest optimism.
Make interoperability a shared priority when evaluating solutions. Emphasize deep EHR integration, structured outputs, and secure data handling. AI should simplify the environment, not fragment it further.
Invite clinicians into the evaluation process. Many physicians are already experimenting responsibly. Involving them in vetting and piloting tools improves fit and builds trust. Then provide safe, sanctioned options for high-friction tasks. If clinicians are using AI to summarize external PDFs or synthesize patient histories, offer compliant, enterprise-supported tools that address that need directly.
Finally, create clear guidance instead of barriers. Establish simpler governance frameworks that clarify appropriate use without discouraging innovation. The goal is enablement, not restriction.