What is AI regulation?
The AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
AI regulation comprises legally binding rules governing how AI is developed, supplied or used. It is different from voluntary guidance, ethical principles or an organization’s internal policy, although those can help implement legal duties. The applicable rules depend on location, function and the roles of the organizations involved.
Some laws address AI broadly. Others apply because of the activity: processing personal information, making an employment decision or providing a medical device. A product does not sit outside existing law simply because it uses a new model.
The European Union’s AI Act combines prohibitions, obligations for specified high-risk uses, transparency provisions and rules for general-purpose models. It operates alongside other regimes, including data protection and medical-device law. Its implementation dates have changed, so organizations should use current official guidance rather than rely on an older summary.
In the United States, health-related AI can fall within several distinct frameworks. FDA oversight depends on whether a software function meets the relevant medical-device definition and exclusions. HTI-1 addresses certain decision-support transparency and risk-management requirements through health IT certification. CMS-0057-F addresses specified payer interoperability and prior-authorization processes. None is a general approval system for every AI application used in a hospital.
Guidance documents also need to be described accurately. FDA guidance explains the agency’s current thinking; the statutory and regulatory provisions establish the underlying legal requirements. NIST’s AI Risk Management Framework is voluntary, not a federal AI licensing law.
Regulatory review and local evaluation answer related but different questions. Authorization for an intended use does not establish that a tool fits every organization’s patients, data and workflow. Conversely, promising local performance does not remove an applicable legal obligation. Continuing monitoring and appropriately controlled changes matter after initial deployment.
For a health organization procuring an assistant, begin by specifying its functions: reading, drafting, recommending or taking action. Then identify the relevant legal roles and requirements, rather than asking whether the product is generically “AI compliant.” Smaller organizations may need shared expertise and assurance resources to do this well. Regulation is one part of protecting patients and the public; a working service also needs clear ownership, evidence and a route to correct harm. This overview does not determine the legal status of an individual product.
Sources: FDA — Clinical Decision Support Software (January 2026); ASTP/ONC — Decision Support Interventions Certification Companion Guide; CMS — Interoperability and Prior Authorization Final Rule CMS-0057-F; European Union — Regulation (EU) 2024/1689, consolidated text; Panch and colleagues — A Distributed Approach to the Regulation of Clinical AI (2022)
To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy
Lead with AI in Health Care
The AI Pragmatist is a monthly newsletter from ALA faculty, with new explainers & essays from the Advanced Learning Academy