AI in Health Care Needs More Than a Hospital Signature

Hospitals are using artificial intelligence to make serious choices about patient care. These tools can help with diagnoses, treatment plans, and who gets seen first. Many health systems check that the AI works well and follows the law. They also watch it after it goes live. But there is a bigger question that often gets skipped. Just because a hospital approves an AI tool, does that mean the public has agreed to let it shape lives?

Institutional approval is not the same as public authorization. A hospital board can sign off on a system. That does not automatically make it fair or right for everyone affected. Sometimes no one takes responsibility for explaining why the AI was chosen or who it might leave out. This is called an unclaimed decision. It happens when a choice about purpose, limits, who is included, or what matters most has no clear owner who must answer to the public.

A proposed framework offers a three-part way to think about this. First, procedural authorization means the process for approving AI is open and fair. Second, substantive justification means the reasons for using the tool hold up under scrutiny. Third, relational renewal means checking back with patients and communities over time. There are also three minimum rules that can be audited. Everyone should be able to see that the AI is deployed. People affected by it should give input before the rules are locked in. And there must be a way to review the decision later, triggered by someone outside the team that built or launched it.

Finally, there is a five-question test for public authorization. It asks about the purpose of the AI. It looks at how benefits and harms are distributed. It considers who might be excluded. It checks whether people have real choices and ways to challenge decisions. These questions push health leaders to think beyond technical checks. They ask who gains, who loses, and who gets to speak up. That is how legitimacy becomes a real requirement, not just an afterthought.