Clinical Trials’ AI Focus Is Shifting From Finding Drugs to Finding People and Building Trust

As discussion of AI in biomedicine gradually moves from “can it design new molecules” to “can it make clinical trials more feasible,” the questions are also becoming closer to patients. A clinical trial is not simply data engineering, but a high-risk system woven together by physicians, participants, ethics review, regulatory requirements, and pharmaceutical company decisions. Any algorithm that enters it must face tests of evidence, transparency, and trust.

The ACT Brief published by Applied Clinical Trials Online on July 9 places three topics side by side: multimodal AI system architecture, trust-based trial recruitment, and AI transformation at large pharmaceutical companies. Because the public summary does not provide full technical details, company names, or specific clinical data, the item reads more like an industry signal: AI’s role in clinical development is expanding from point tools into workflow infrastructure.

In the context of clinical trials, so-called multimodal AI usually means systems that do not only read structured tables, but may also integrate medical-record text, images, laboratory values, trial documents, and participant-reported outcomes. If such architectures operate properly, they may help screen eligible patients, detect data anomalies, organize trial documents, or help research teams grasp safety signals more quickly. But they can also amplify differences among data sources, accumulated bias, and issues of model explainability.

Participant recruitment is especially a setting where AI is most easily expected to help, and also most likely to cross boundaries. Clinical trials have long faced difficulties such as insufficient recruitment, complex eligibility criteria, and inadequate population representativeness. AI can be used to compare inclusion and exclusion criteria, help identify potential candidates, and even improve communication of trial information. However, if patients do not know how their data are being used, or if a model systematically excludes certain groups, efficiency may come at the cost of trust.

Therefore, “trust-based recruitment” is not a public-relations phrase, but a set of concrete requirements: data use needs clear consent and governance mechanisms; screening recommendations need to be reviewable by researchers; recruitment messages must not mislead patients’ expectations about efficacy; and system performance should be tested across different diseases, regions, and populations. For regulators and ethics committees, the question is not only whether AI has increased recruitment speed, but whether it has changed who can see trials, who is invited to participate, and who is excluded.

AI transformation at large pharmaceutical companies represents another layer of pressure. Generative AI and machine learning have already been introduced into literature review, target assessment, trial design, medical writing, and operations management, but the core of clinical development remains evidence that is reproducible, auditable, and acceptable to regulators. If AI stays only at the level of improving internal efficiency, its impact is limited. If it begins to influence trial design, patient stratification, or safety monitoring, companies must establish stricter systems for validation, accountability, and human review.

**Background Context**

Recent discussion of biomedical AI is no longer centered only on the imagination of “AI drug discovery.” Whether candidate molecules can reach approval still depends on whether human trials can demonstrate efficacy and safety. The speed, quality, and fairness of clinical trials themselves are becoming AI’s next, more difficult testing ground. Although the keywords raised by the ACT Brief are limited by the summary, together they point to the same issue: if biomedical AI is to move from demonstrating capability to changing medical R&D, it must first learn to accept constraints in clinical settings.