Cato Institute urges FDA overhaul to keep pace with AI

A new policy analysis from the Cato Institute is calling on FDA to modernize its regulatory framework for artificial intelligence (AI) in healthcare, arguing that current rules are slowing innovation and discouraging companies from developing high-impact medical tools. In the 4/13 blog post, authors Christopher Gardner and Kevin T. Frazier contend that FDA’s existing medical device review process — originally established under the 1976 Medical Device Amendments — imposes significant financial and time burdens that are ill-suited to the rapid development cycles of AI software.

The authors argue that bringing AI-enabled medical products to market can cost companies anywhere from hundreds of thousands to tens of millions of dollars, depending on device classification, with review timelines stretching up to a year. Given that AI models can evolve in a matter of months, such delays risk rendering products outdated before they receive clearance or approval, they said.

As a result, many developers are choosing to avoid FDA oversight altogether by focusing on applications that fall outside the agency’s jurisdiction, such as administrative tools or general wellness software. The authors say this trend limits the development of more advanced clinical applications, including tools that could assist physicians with diagnosis, reduce medical errors, and improve care coordination.

Central to the critique is FDA’s framework for Software as a Medical Device (SaMD), which classifies software intended for diagnosing or treating disease as a regulated medical device. According to the authors, this definition discourages companies from optimizing general-purpose AI systems for healthcare use, for fear of triggering costly regulatory requirements.

The report points to examples such as general-purpose AI chatbots, including those developed by OpenAI, which have the technical capability to support clinical decision-making but are often constrained to non-medical or “wellness” use cases to avoid FDA scrutiny.

The authors also highlight inconsistencies in enforcement, noting that some AI-driven “symptom checker” tools are allowed to operate under enforcement discretion despite meeting the agency’s definition of medical devices. They argue that this selective approach contributes to regulatory uncertainty for developers and investors.

To address these challenges, the Cato analysis recommends reforms to the FDA’s SaMD guidance, including a system of conditional enforcement discretion. Under such a model, AI developers could bring products to market more quickly in exchange for meeting transparency requirements and reporting adverse events, similar to existing medical device reporting systems.

The authors acknowledge that the FDA has taken recent steps to adapt, including allowing predetermined change control plans for AI software and expanding flexibility for certain clinical decision support tools. However, they argue that these changes rely largely on guidance rather than formal rulemaking, leaving them vulnerable to reversal by future administrations.

They also warn that the agency’s reliance on a product’s “intended use” may not adequately address the rise of general-purpose AI systems that are increasingly being used by patients for health-related information, even if not explicitly marketed for that purpose.

Drawing a parallel to the slow adoption of 3D printing in healthcare, the authors caution that regulatory uncertainty could similarly delay the integration of AI into clinical practice. They urge the agency to pursue more durable policy changes and to create clearer pathways for AI tools that reflect how the technology is already being used.

“The FDA is taking steps in the right direction,” the authors conclude, “but broader reform is needed to ensure that innovation in AI-driven healthcare is not stifled by outdated regulatory frameworks.”