AI-Enabled Workarounds Pose Regulatory Challenge
Commentary from technologist Toby Rogers suggests that AI is enabling the creation of chemical workarounds that complicate the regulation of biologics and toxic substances. The analysis notes that regulatory bodies are currently lagging behind the pace of these AI-driven innovations.
- The U.S. Food and Drug Administration (FDA) is actively developing regulatory frameworks for AI in drug manufacturing through initiatives like the Framework for Regulatory Advanced Manufacturing Evaluation (FRAME). Concurrently, the Environmental Protection Agency (EPA) is developing tools such as an "AI Chemist Assistant" to accelerate chemical safety reviews under the Toxic Substances Control Act. - A primary regulatory hurdle is the "black box" nature of some AI systems; deep learning models often produce results without a clear, traceable rationale, which conflicts with the need for explainable and reproducible data required by bodies like the FDA and European Medicines Agency (EMA). - Generative AI is being used for the *de novo* design of biologics, including creating entirely new antibodies and proteins from scratch. In the gene therapy space, AI models are being used to design and optimize Adeno-Associated Virus (AAV) capsids for improved cell-specific targeting and to evade host immune responses. - The speed of AI-driven discovery is a key factor; for example, Insilico Medicine used its generative AI platform to move a novel anti-fibrotic drug candidate from target identification into Phase I clinical trials in just 30 months. - While generative AI can propose novel molecules, a significant bottleneck is synthetic feasibility, as many AI-suggested chemical structures are difficult or impractical to manufacture, slowing lab adoption. - The reliability of AI outputs is a major concern, as models trained on incomplete, biased, or low-quality datasets can generate flawed conclusions. This issue of data integrity is a critical challenge for validation in GMP environments. - AI is enabling the creation of entirely new functional biological entities. Scientists have successfully used AI to design and build bacteriophages—viruses that infect bacteria—capable of targeting specific strains, including antibiotic-resistant ones. - A key challenge in implementing AI within the heavily regulated chemical industry is the prevalence of unstructured data scattered across legacy systems and PDFs. Poor data quality can lead to AI "hallucinations," where the model generates plausible but incorrect outputs, posing a risk to compliance and safety.
