News

Artificial intelligence (AI) has the potential to transform every step of the drug development, from early discovery to clinical trials, and pharmaceutical and biotech companies are eager to harness its capabilities to bring life-saving medicines to market faster and more affordably. How federal agencies regulate the use of AI in this context will help determine if that promise is realized.

According to Duke Law professor Arti K. Rai, there is much to be gained by looking beyond national borders as AI policy in drug development takes shape. Rai, an expert in innovation policy, has been studying how the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) approach oversight of AI in two of the world’s largest pharmaceutical markets and what each can learn from the other.

Studying EMA regulation, Rai said, can “expand the horizons of what one can think about in terms of how our domestic regulatory regime is structured.” Rai, the Elvin R. Latty Distinguished Professor of Law at Duke Law School and co-director of Duke’s Center for Innovation Policy explores these themes in a new paper comparing the FDA’s more flexible, case-by-case approach with the EMA’s structured, risk-tiered regulatory framework.

Comparing these two regulatory agencies also highlights challenges and opportunities for companies hoping to take advantage of AI-driven innovations, she writes in the paper co-authored with European legal scholars who were equally interested in studying FDA processes.

Due to their different cultural, political, and economic environments, the agencies have long taken different approaches to drug development: the FDA’s approach is characterized by dialogue with industry and numerous accelerated approval pathways, while the EMA generally takes a more conservative approach. 

That’s reflected in their approach to AI, Rai says. Both American and European regulators have signaled their support for incorporating AI. In 2024 the EMA released a reflection paper on AI in drug development that “reflects the European Union (EU)’s broader strategy of implementing comprehensive technological oversight while maintaining sector-specific requirements for pharmaceutical development,” the authors write. The regulatory architecture set out in the EMA paper takes a tiered, risk-based approach with less oversight for drug discovery applications that have minimal direct patient impact, and enhanced regulatory scrutiny for clinical trials.

The FDA, by contrast, has approached the regulation of drug development on a more flexible and context-specific basis, characterized by informal guidance and individualized evaluation – what the authors call “artisanal regulation.” The Trump administration has explicitly called for AI leadership in all sectors of the economy, and the FDA released a discussion paper on the use of AI in drug development, followed by early guidance on the use of AI to support regulatory decision-making and principles for good AI practice that will also “help cultivate future growth.” 

Different approaches, a common goal

While there hasn’t yet been a drug approved in either the US or the EU that primarily relied on AI in its development, differences in the way the two agencies view AI’s role are beginning to emerge, Rai said. For example, the FDA hasn’t emphasized scrutiny of AI’s use at any stage prior to the administration of a drug candidate to humans in clinical trials. 

“I think that’s a very deliberate choice on their part,” said Rai. “I'm a little more skeptical. I think that there should be really careful preclinical analysis of what is going to go into the human.” 

The agencies also take different approaches to the “black box” nature of AI models, in which it is impossible to understand how a model arrived at its results. 

“When there are few explanatory criteria associated with the model, the EMA might think it's much more important for the regulator to be heavily involved in the development process,” said Rai. “The US has not expressed such a clear preference for less ‘black boxiness’ or for what are sometimes called ‘clear box’ models, whereas the EMA has definitely expressed that preference.” 

Nonetheless, Rai believes the two agencies have much to learn from one another: the FDA could benefit from the EMA’s clear standards and processes, since artisanal regulation requires numerous personnel, while the EMA could take a page from the FDA’s individualized approach, which is especially helpful for small and medium-sized firms that lack the experience or resources of big pharmaceutical companies. 

“In general, smaller firms are the places where the most interesting earlier stage research is done, and I think that's going to be even more the case as a consequence of AI,” said Rai. “The US could make clear to the rest of the world exactly how they engage with small entities to help them along. That would be particularly useful for the EU, which also wants to prioritize small firms.”

Recognizing their mutual interest in developing medicines that could help humanity on a global scale, the FDA and the EMA have released a joint memo laying out areas of collaboration and common principles to guide AI use in drug development. “As the use of AI in drug development evolves, so too must good practice and consensus standards,” the memo reads. “Strong partnerships with international public health partners will be crucial to empower stakeholders to advance responsible innovations in this area.”

"If regulators can find the right balance between encouraging innovation and ensuring safety and trust in the process, it could unlock tremendous benefits for human health,” Rai said.

“In contrast to other use cases for AI, the benefits to society could be really significant,” she said. “If the benefits are likely to be higher, then maybe one should encourage AI use more.”