Healthcare AI Is Deployed Nationwide. Governance Isn’t Ready

Healthcare AI is already changing how care gets delivered. The real question is whether the systems around it are ready.

That shift is starting to show up beyond hospitals. When the U.S. Senate titled a hearing “Less Hype, More Help,” the conversation has clearly moved on. Earlier this month, I testified before the Senate Subcommittee on Science, Manufacturing and Competitiveness, where the focus reflected that change. My written testimony reflects my experience deploying these systems in real clinical settings as Chief Innovation Officer at Rad AI, as well as my perspective as a patient and caregiver.

For years, the dominant question was whether AI would replace doctors. A decade of debate, and we were mostly arguing about the wrong thing.

The real story is not elimination; it is integration. AI is already changing how clinical work gets done – redistributing tasks, compressing certain types of work and elevating others – inside some of the most complex, highly regulated systems in the economy. Whether that change strengthens patient care or simply adds noise depends on decisions being made right now in hospitals, regulatory agencies and Congress.

I know this personally. It took 10 years and a string of dismissed symptoms before I was diagnosed with multiple sclerosis. I did not fit the typical patient profile. My records were siloed. No system flagged the patterns of symptoms. No one was looking across the full patient picture. This is the very gap AI is meant to close, if we deploy it responsibly. At some point, every one of us will be a patient, a caregiver, or both.

Sharing the table during the Senate hearing were a robotics company deploying humanoid robots in Amazon warehouses, a Siemens executive overseeing AI integration in U.S. shipyards and a Brookings fellow tracking how these shifts are playing out differently region by region. As Siemens executive Brittany Ng put it, “The issue is not AI coming for industrial jobs. It’s a shortage of skilled workers amid increasing production complexity.” Healthcare is no different.

In shipbuilding, Siemens is simulating entire vessels digitally before a single piece of steel is cut, compressing design cycles from months to hours. A new facility in Fort Worth, Texas, modeled the full production floor in a virtual environment before breaking ground and still created hundreds of jobs. In warehouse logistics, Agility Robotics pointed to Amazon, which deployed more than 1 million robots over 13 years while adding 1.4 million human employees in the same period. Manufacturing today has 400,000 open positions it cannot fill. The issue is not automation taking jobs, but a shortage of workers to do them. Healthcare is no different.

‘Cleared For Market’ Does Not Mean Ready For 2 A.M.

The U.S. already has a rigorous regulatory framework for medical technology. The Food and Drug Administration has cleared over 1,400 AI-enabled medical devices, and existing mechanisms allow systems to be updated while still being monitored. Innovation has not ground to a halt.

But healthcare AI does not behave like a static piece of equipment. Performance can drift. Clinical workflows change. Data inputs evolve. These systems interact with human judgment in ways that vary depending on the setting, the patient population and the time of day. Regulatory approval before a product reaches patients remains necessary. It is simply not sufficient on its own.

Major professional societies have recently made this point explicitly: the future of healthcare AI regulation must include robust monitoring after a system is deployed and ongoing evaluation of how it performs in the real world. This reflects a basic reality: dynamic systems require ongoing oversight.

Benchmarks matter. What matters more is what happens at 2 a.m. in a community hospital when a real patient is waiting for an answer.

We Are Scaling Systems We Cannot Yet Measure

What has surprised me most in deploying these systems over the past five years was not the technology; it was the gap between detection and action. A patient can be sitting in a waiting room while a blood clot is already visible on their scan, and AI can flag it in minutes, but the flag is not the outcome. Was it seen? Was it acted on? Was it acted on in time?

We obsess over whether the model is right while ignoring whether the system responds at all. Without infrastructure to measure what happens after the flag, we are deploying blind. That is not a technology problem; it is a measurement problem, and right now we are treating it like someone else’s job. Congress can play a direct role here: supporting privacy-preserving national datasets used to validate AI performance, encouraging standards for monitoring systems over their full lifespan and strengthening the technical frameworks that allow AI tools to be evaluated consistently across different health systems. Right now, much of that oversight happens in isolation, hospital by hospital.

Radiology attrition rates have almost doubled over the past decade while imaging volume is projected to rise 26% over the next 30 years. The workforce is not disappearing; it is overwhelmed. Radiology was simply the first specialty in the crosshairs.

Governance Is Not The Brake. It Is The Engine

Healthcare runs on trust in a way that most industries simply do not.

If AI systems are perceived as opaque, unmonitored or unaccountable, adoption will stall regardless of how capable the underlying technology is. Responsible governance is not a brake on innovation in healthcare — it is what makes innovation durable.

Recent federal discussions have included proposals to expand access to government data to accelerate AI development. In healthcare, the more immediate opportunity may lie on the validation side. Carefully designed, de-identified datasets used for post-deployment evaluation could build trust as effectively as training data accelerates development.

The next phase of healthcare AI will not be defined by predictions; it will be defined by how well the field measures performance, monitors safety over time and ensures accountability as these systems scale.

Some of the most respected voices in AI predicted radiologists would be obsolete. Others said the same about human drivers. Radiologists are still driving themselves to work. The more important question is whether the tools waiting for them are governed in a way that actually strengthens care.

The original version of this blog originally appeared on forbes.com and is reposted with permission.