EU AI Act Could ‘Set Back’ Benefits of AI in Drug Development If Applied to R&D
The EU’s Artificial Intelligence (AI) Act was formally adopted last year, with many of its provisions expected to apply this year or next. For pharmaceutical and biotech companies, however, ambiguities in the legislative text may raise concerns around how extensively the Act will apply to drug R&D activities.
Concerns from Industry Experts
This potential challenge was highlighted by Stephen Reese, co-chair of the healthcare & life sciences sector group and co-head of the intellectual property group at the international law firm Clifford Chance.
The key question for drug developers in relation to the AI Act is how much it will impact drug development activities. “Some in the industry are expressing concerns that AI legislation is more far-reaching than potentially necessary, as it could extend deep into aspects of the early stage of the drug lifecycle,” Reese stated.
He emphasized the importance of assessing where AI systems are being used for R&D, which could include theoretical drug discovery analyses, where companies may use AI to process data and gain insights into disease pathways.
Regulatory Ambiguities and Exemptions
The AI Act covers all sectors, and although AI tools used only for R&D are theoretically exempt from the legislation’s strict rules, it remains unclear how this exemption would work in practice.
While the AI Act stipulates that the company marketing or trademarking an AI product is responsible for ensuring compliance, researchers may still need to review their license terms, contracts, warranties, and liabilities when using AI tools.
Reese noted that the AI Act, if applied to drug discovery activities, could make the UK a more attractive destination for research than the EU.
AI in Clinical Trials and Bias Risks
AI could also be effective in clinical trials, for example, to find the right patient population groups to target for recruitment. In this case, companies would use AI to process large amounts of data to identify suitable substances and the right groups of people for clinical trials.
However, Reese pointed out that there is a risk of unexpected bias when using AI to select participants for clinical trials. This could skew results compared to a more randomized recruitment process.
Regulatory Sandboxes: A Double-Edged Sword
One aspect of the AI Act that has been largely welcomed across industries, including pharma, is the introduction of “regulatory sandboxes,” which provide controlled environments for developers to test their products under regulatory supervision.
However, there are “question marks” over what can be included in a sandbox regime. Sandboxing could be employed to explore initial drug discovery work or the use of digital twins for testing a product’s efficacy and safety profile.
Digital twins are virtual replicas of biological systems designed to simulate the effects of drugs on real-life patients. While AI-enabled digital twins could introduce regulatory complexities if widely adopted, Reese mentioned that the industry is still in the early stages of using digital twins.
Conclusion
Drug regulators will expect stronger proof cases before incorporating digital twin technology into assessing medicines. Established pre-clinical and clinical study methods are unlikely to be replaced by digital twin technology in the near future.
This article is part of a series that discusses the implications of the EU AI Act on drug development, including unintended data ownership and intellectual property challenges.
