States Split on Rules for AI in Medicine

Artificial intelligence is being used more and more in hospitals and doctors' offices across the United States. But there is no single national law that explains exactly how that technology must be used safely. That means each state is starting to write its own rules, and a top health AI expert is warning that hospitals need to get ready for a very messy system.

Brian Anderson is the CEO of the Coalition for Health AI, a group that works to make AI in medicine safer and more consistent. He spoke at a health AI event in New York City. He said hospitals should expect 'a patchwork set of regulatory frameworks' over the next one to two years. That means different states will have different sets of rules, like a quilt made of mismatched pieces.

So far, the push for state rules is moving fast. More than 280 bills about health AI were introduced in state legislatures in just the first half of 2026. By the end of July, 33 of those bills had been signed into law across 22 states. Anderson said the process is already moving and cannot be stopped — 'That train has left the station,' he explained.

One big concern is the use of AI systems that can act more on their own, called 'agentic' systems. These systems can carry out tasks without a human making every single decision. Hospitals are very worried about using this kind of AI in situations where mistakes could seriously hurt patients. Anderson said health systems are 'deeply concerned' about these high-risk uses.

A key question is how accurate AI needs to be before it can be trusted with important medical tasks. Anderson gave the example of an AI tool that predicts a dangerous illness called sepsis. He asked whether 95 percent accuracy is good enough — or 80 percent, or even 70 percent. 'We don't know the answer to these questions,' he said honestly.

One state is already testing a new idea. Utah is working with a company called Doctronic, whose AI can help renew existing prescriptions. Renewing a prescription is different from refilling one — it usually requires a licensed doctor to approve it. Right now, a human medical professional still checks and approves every request in the program.

The goal of the Utah program is to help patients who have trouble seeing a doctor just to get a prescription renewed. This is especially a problem in areas where there are not enough doctors. However, Utah's Medical Licensing Board recommended pausing the program in April. Board members argued that renewing a prescription should include a fresh check of the patient's health, and they raised concerns about safety.

Anderson agreed that even this lower-risk use of AI carries some danger. A patient's blood test results might show something that needs a doctor's attention. The AI might miss a sign that a medication needs to be changed or stopped. 'So it's not without risk,' Anderson said.

The bigger worry is what happens if every state ends up with completely different rules. Health systems and tech companies that work across the country would have to follow up to 50 different sets of laws. Anderson warned that companies might decide it is simply not worth the effort to operate in smaller states if the rules are too different and too hard to follow.

The Coalition for Health AI is working with states and other groups to encourage them to agree on similar rules. Anderson said everyone understands that too much difference will slow things down. His hope is that the federal government will eventually step in and create one clear set of guidelines. 'My hope certainly is that at the federal level, there is some overarching set of frameworks and regulatory guidance that's given,' he said.