Health systems are spending more on artificial intelligence than ever before. The tools are more capable, the vendor pipelines are fuller and the board-level pressure to move has rarely been more intense. Yet across organizations, roles and geographies, the leaders closest to AI implementation are raising alarms — not about the technology itself, but about the decisions health systems keep making around it.
In responses gathered by Becker’s, more than two dozen CMIOs, CMOs, CNOs and health IT executives identified what they consider the most dangerous trends in health IT.
1. Treating governance as a checkbox
The most consistent concern across respondents was not a technology failure. It was an organizational one: approaching AI governance as a tactical foundation rather than a clinical function.
“We’re deploying AI faster than we’re learning how to govern it,” said Hasan Ahmad, DO, vice president and associate CMIO of Parkview Health. “Too many organizations treat governance as a procurement checkbox — something you do once before go-live. We validate a predictive model, turn it on, and then let it run for years. Meanwhile the patient population shifts, documentation patterns change, and performance quietly drifts. Nobody’s watching.”
Dr. Ahmad said the inability to turn functionality off contributes to drift. At Parkview Health, his team has removed more than 100,000 interruptive alerts per month, a figure that points to how much noise accumulates in clinical environments before anyone intervenes.
“Deimplementation is a clinical skill,” he said. “Most organizations haven’t built that muscle yet. If we drop AI into environments already saturated with noise clinicians have learned to ignore, they’ll ignore the AI too.”
Toyosi Olutade, MD, chief medical officer at UnityPoint Health-Quad Cities in Moline, Ill., framed the governance gap in terms of scale. When AI is wrong, it is wrong everywhere at once.
“The greatest risk is that we scale error just as quickly as we scale value,” Dr. Olutade said. “If the underlying data are flawed, models drift, or clinicians begin to rely on technology at the expense of clinical judgment, the impact can spread rapidly across an entire health system.”
2. Automating flawed processes
A second pattern runs deeper than governance. Several respondents argued that the fundamental error is not how health systems manage AI once it is deployed, but what they ask it to do in the first place.
Enitza George, MD, associate professor and chief population health officer at SUNY Downstate Health Sciences University in New York City, said technology scales whatever system it is given, including its inefficiencies and its flawed assumptions.
“Before asking whether a process can be automated, we need to ask whether that process should exist in its current form at all,” Dr. George said. “Not evaluating the systems prior to automation is dangerous because once you automate a flawed process, you don’t just preserve the flaw, you scale it. You make it faster, harder to see, and potentially much harder to undo. In healthcare, the consequence isn’t just wasted money or inefficiency. It can affect access, equity, clinician behavior, and ultimately patient outcomes and safety.”
Shane Borkowsky, MD, associate chief health information officer at UI Health in Chicago, identified the same problem inside the EHR. Inaccuracies in clinical documentation — missed dictations, copy-forward errors, uncorrected diagnoses — become the raw material AI tools learn from and act on.
“When an incorrect diagnosis or result or document goes unchecked and is now taken as fact, the resultant tools, including documentation, summaries, orders, patient education, may propagate the errors,” Dr. Borkowsky said. “When we rely on the document without reviewing the content, it is easy to miss small errors in attribution or in content.”
3. Building AI without clinicians in the room
AI tools are being designed and purchased without the people who will use them having meaningful input into how they are built. Terry McDonnell, DNP, RN, senior vice president and chief nurse executive at Duke University Health System in Durham, N.C., has seen the benefits of including end users early.
“The best solutions are driven by the people who are closest to the work,” Ms. McDonnell said. “Having clinical advisors or an advisory board is different from co-development. There are a lot of companies out there that are very well-intended but not developing with a co-development viewpoint.”
The consequence of that gap is not just poor adoption. It is waste — and in an industry where margins are thin and AI budgets are growing, that distinction matters.
“When the wrong solutions are brought forward, that creates waste, and I think we can all agree that we can’t afford to introduce waste into our already complex systems,” Ms. McDonnell said.
Tomi Kolade, MBBS, assistant CMIO at UTHealth Houston, focused specifically on agentic AI as the place where clinician exclusion becomes most dangerous.
“A poorly designed screen frustrates a physician,” Dr. Kolade said. “A poorly designed agent acts on the wrong assumption: routing a message that needed a physician’s eyes, closing a loop that wasn’t actually closed, executing a workflow with no one watching until the outcome is already downstream.”
“The future of the EHR should not be AI designed for clinicians,” he said. “It should be AI designed with clinicians.”
4. Adding complexity instead of reducing it
AI is frequently making clinicians’ work harder, the opposite of its intention. More alerts, more inbox messages, more documentation requirements, each tool added in the name of efficiency compounds the cognitive load on the people delivering care.
“Healthcare is deploying more technologies than ever, yet many clinicians feel less productive and more burdened,” said Salim Saiyed, MD, CMIO of UT Health Austin. “We are frequently adding more complexity without adding clinical value. The biggest danger is adding to cognitive overload. If we are not intentional, technology can actually amplify the problem by generating more information for clinicians to process rather than reducing the burden.”
Anitra Williams, DNP, area chief nurse executive and interim COO at Kaiser Permanente Central Valley, framed the standard for innovation differently.
“Innovation should never be measured by how quickly we implement a new tool, but by whether it measurably improved the experience and outcomes of patients and the people caring for them,” Ms. Williams said.
Hetal Rupani, senior director of business intelligence and analytics at Johns Hopkins Medicine in Baltimore, sees further problems with the blurred line between what EHR vendors enable and what health systems must own.
“EHR platforms are foundational, but the responsibility for making technology scalable, usable, and clinically meaningful sits with health systems,” Ms. Rupani said. “When that distinction is not clear, organizations can end up with expensive systems that are technically live but operationally underperforming.”
5. Chasing the technology instead of the problem
Health systems are selecting AI tools because they are available, compelling and on-trend, not because they address a defined clinical or operational need.
Mohamed Rami Nakeshbandi, MD, vice president and CMO at Downstate Health Sciences University in New York City, described the dynamic that makes this so difficult to resist. Vendors are persistent, promises are large, and the frameworks for objective evaluation are thin.
“Hospitals and leaders are being overwhelmed by AI and other technology vendors knocking on our doors every day, each promising to improve quality, efficiency, and revenue cycle,” Dr. Nakeshbandi said. “But we have very limited standardized ways to objectively evaluate which solutions truly meet our clinical need, deliver meaningful value, and have the real return on investment. The danger is that we end up adopting a technology because it is available and compelling rather than because it is proven to improve efficient care.”
Deepa Velayadikot, MD, medical director of care coordination and regional medical director of hospital medicine at Cooper University Hospital in Camden, N.J., called for a more deliberate standard, what she termed “relevant AI.”
“AI should solve a specific problem, improve care delivery, enhance efficiency, or support better decision-making, rather than being adopted simply because it is the latest trend,” Dr. Velayadikot said. “Without thoughtful evaluation, strong governance, and alignment with organizational goals, AI risks becoming an expensive buzzword rather than a meaningful tool for healthcare transformation.”
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
