On 14 January 2026 the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) jointly published a set of ten guiding principles on good artificial intelligence practice in drug development. The common framework, produced after months of collaboration between the two regulators, is intended to guide the generation of evidence by AI at each stage of drug research and development.
The orientation of the ten principles
According to public information, the ten principles cover dimensions from “human-centred design” to “clear governance and accountability”. Their core orientation can be summarised in three points.
First, AI’s role is to assist, not replace. The principles stress the ultimate role of humans in decision-making; AI output should support, not supplant, professional judgement.
Second, lifecycle management. The principles require traceable records at every stage — data, model, application and interpretation of results — rather than focusing only on final output.
Third, proportionality to risk. The principles set different evidence and validation requirements according to the risk level of the AI use, avoiding a one-size-fits-all approach.
Why the regulators acted jointly
The joint move by the FDA and EMA reflects the globalisation of drug development. The same drug is often filed in several jurisdictions at once; if each regulator’s AI requirements conflicted, companies would face duplicative validation costs. The value of common principles is providing a reusable “minimum consensus”.
This also continues the recent path in AI governance of “soft law first, hard law later”: regulators first issue principled guidance, and only once practice matures do binding rules follow.
Implications for pharmaceutical AI companies
First, explainability of evidence generation. Where AI participates in preclinical research or data analysis, its methods, data sources and validation process must be understandable to reviewers. See our note on AI application ethics and security guidelines.
Second, data governance moves upstream. Drug development involves large volumes of subject and patient data, and the compliance requirements for outbound data are especially strict. See the PI outbound certification measures and our note on the obligations of major personal information processors.
Third, managing differences across two jurisdictions. Although the FDA and EMA published common principles, their review processes and high-risk classifications still differ. Companies serving the Chinese, US and European markets at once need a mappable multi-jurisdiction compliance plan; see our multi-jurisdiction service.
Fourth, interface with domestic regulation. China’s drug regulator is also advancing “AI plus drug regulation”, and the direction of regulatory science converges with that of overseas agencies; see our note on the opinion on AI plus drug regulation.
Practical recommendations
Pharmaceutical AI companies can use these principles as a self-check framework for internal R&D processes: mark AI use and risk level at project initiation, build an authorisation-chain ledger at the data stage, retain validation records at the model stage, and set human review nodes at the output stage. Where cross-border R&D cooperation is involved, the outbound data route should be assessed in advance; see our data compliance service and cross-border data service.
Regulatory science’s acceptance of AI is an important opportunity window for the pharmaceutical industry. But the ticket through that window is not model performance — it is governance capability that is verifiable, explainable and traceable.
This article is based on the ten guiding principles on good AI practice in drug development jointly published by the FDA and EMA, and public reporting. It is for general reference only and does not constitute legal advice.