Between September 2024 and January 2026, FDA and EMA published their first formal regulatory frameworks for AI in drug development, culminating in joint principles that define expectations for any organization using AI to generate or support clinical trial data intended for regulatory review.
FDA's seven-step credibility assessment framework, released in January 2025, requires sponsors to define the specific context of use for each AI model, assess its risk based on influence and consequence, and document validation proportionate to that risk. The framework applies across the drug lifecycle, from nonclinical through post-marketing, but excludes AI used solely for discovery or internal operations.
EMA's September 2024 Reflection Paper places direct accountability on sponsors to demonstrate AI systems are fit for purpose and comply with EU law and GxP standards. Where full explainability isn't possible, EMA requires interpretability: evidence that human oversight can identify problems and intervene when models underperform.
The frameworks converge with ICH E6(R3), adopted by FDA in September 2025, and the EU AI Act, which began phased enforcement in February 2025. Not all clinical AI is automatically high-risk under the Act, classification depends on intended use and whether the system materially affects health decisions. For site staff, the practical takeaway is that AI used anywhere in the regulatory evidence chain now faces mutually reinforcing expectations from multiple authorities simultaneously.