The UK's medicines regulator has announced plans to create a controlled testing environment for artificial intelligence tools in pharmaceutical development, in what represents a formal step towards integrating AI into the drug approvals process.

The programme, funded by the government body responsible for modernising regulatory infrastructure, will allow AI systems to be assessed for their ability to predict how medicines are absorbed and broken down by the body, and to identify potential harms at an early stage. Up to five different AI approaches will be tested in the first phase. The regulator will begin working with partners from industry and academia from this summer.

The stated aim is not simply to explore what AI can do, but to assess whether it can be trusted. Regulators want to know how reliable these tools are when applied to decisions about the safety of new medicines, and the findings are expected to contribute to formal guidance on their use in the development pipeline. That distinction matters. AI has been applied in drug discovery in various forms for several years, but its formal incorporation into regulatory decision-making has remained limited, partly because the evidence base for its reliability has been thin. Testing tools in a controlled environment before drawing any conclusions is, in that respect, the appropriate place to start.

The programme will also examine how clinical data can be used more effectively to understand how medicines perform across different population groups. This reflects a longstanding concern in pharmaceutical research about whether approved medicines are adequately tested across the range of patients who will ultimately use them. If AI tools can help extract more meaningful information from existing clinical datasets, that would be a practical benefit independent of any broader efficiency gains in the approvals process.

Participation from both industry and academic partners from the outset is notable. Pharmaceutical companies have a direct commercial interest in faster and more predictable regulatory pathways, while academic researchers bring methodological independence. How that balance is managed will be one of the more consequential design decisions in the programme's implementation, though the regulator has not yet detailed the governance arrangements.

The announcement arrives as governments and scientific institutions across the world are committing significant resources to AI in health. An international competition has awarded funding to twenty research teams developing AI-based treatments for a progressive neurological condition, granting them access to a genomic dataset covering nine thousand patients. A joint scientific partnership between Britain and France is applying AI and supercomputing infrastructure to infectious diseases including tuberculosis and malaria, supported by public funding from both governments. Canada, meanwhile, has published a national AI strategy that includes a dedicated programme targeting improved health outcomes, backed by hundreds of millions of dollars in committed spending.

The UK initiative is more modest in scale than some of these, but its regulatory focus gives it a different kind of significance. Most AI health programmes are oriented towards research or clinical deployment. This one is asking a prior question: before AI tools can be used to make consequential decisions about new medicines, what standard of evidence is needed to justify that trust? The answer to that question will matter not only for the UK but for any regulator watching how this plays out.

The regulator has indicated that findings will feed into the development of clearer national guidance on the use of AI in medicines development. What form that guidance takes, and how binding it will be, has not been set out. Guidance that is too permissive risks validating tools before the evidence warrants it; guidance that is too restrictive could slow the adoption of methods that turn out to be genuinely useful. Getting that balance right is the harder task that follows the testing itself.

Whether findings from this programme will influence how other major regulators approach comparable questions is open. The European Medicines Agency and the US Food and Drug Administration are both engaged in their own processes around AI in drug development, and the UK's evidence base, once established, could contribute to that wider conversation. For now, the programme represents a carefully framed start.

‍