FDA and EMA set 10 principles

AI in drug development has been moving faster than the rulebook. Companies are already using models to screen compounds, shape trials, monitor safety, and help run manufacturing. But regulators have had a gap — plenty of activity, not enough shared expectations. That changed on January 14, 2026, when the U.S. FDA and the European Medicines Agency published a joint set of 10 principles for good AI practice in the medicines lifecycle. (ema.europa.eu) ### What did the agencies actually do? They did not issue a binding regulation. They published a common principles document — a high-level framework meant to guide how AI should be developed, validated, deployed, and maintained when it is used to generate evidence for drugs and biologics. The scope is broad: nonclinical work, clinical development, post-market monitoring, and manufacturing all sit inside it. (ema.europa.eu) ### Why does this matter now? Because AI has escaped the lab demo phase. Regulators say use of AI across the drug product lifecycle has increased significantly in recent years, and the tools are now touching decisions that feed into quality, efficacy, and safety. Once AI starts shaping evidence for approval or ongoing oversight, “move fast and see what happens” stops being acceptable. (ema.europa.eu) ### So what are the 10 principles? The list starts with the big framing ideas: AI should be human-centric by design, and oversight should scale with risk. Then it gets practical — follow legal, ethical, technical, cybersecurity, scientific, and GxP standards; ensure data quality; keep models fit for purpose; document development and use; monitor performance; manage changes; and (ema.europa.eu)d to know what it was built for, what data shaped it, how well it works, and what happens when it drifts. (fda.gov) ### Why are “principles” useful if they are not rules? Because they tell companies what the regulators are likely to care about before more detailed guidance arrives. That matters a lot in pharma, where global development programs routinely aim at both U.S. and European approvals. A shared FDA-EMA baseline lowers the risk that teams build one AI workflow for one regulator and then discover it is poorly documented, weakly validated, or impossible to defend somewhere else. (ema.europa.eu) ### How does this connect to existing FDA policy? The FDA already moved in this direction with its January 2025 draft guidance on AI used to support regulatory decision-making for drugs and biologics. That draft laid out a risk-based credibility assessment framework tied to the model’s context of use. The new joint principles do not replace that work — they widen it and align it(ema.europa.eu)ared transatlantic philosophy underneath it. (fda.gov) ### What will companies have to do differently? The biggest shift is cultural as much as technical. Drug developers will need tighter documentation, stronger model governance, better data provenance, clearer validation plans, and monitoring after deployment. The catch is that many flashy AI use cases are (fda.gov)hat is exactly the failure mode these principles are trying to head off. (fda.gov) ### Does this mean Europe and the U.S. are fully aligned? Not fully. The EMA says EU-specific guidance is still being developed, building on its 2024 AI reflection paper and other legal frameworks. But the joint publication is still a real step — it signals that future guidance on both sides will rhyme, even if the fine print differs. For global pharma teams, that is a big deal. (ema.europa.eu)) ### Bottom line This is the regulators telling pharma that AI is welcome — but only if it behaves like regulated infrastructure, not a black box. The headline is not “10 new rules.” It is one shared expectation: if AI helps make the evidence, companies must be able to show why the system is trustworthy. (ema.europa.eu)