From Potential to Practical at DIA 2026: Building Trust in AI at Scale

As artificial intelligence becomes more embedded in clinical development, the focus is shifting from capability to trust. At DIA 2026, discussions emphasized that scaling AI requires more than performance. It requires shared frameworks, clear definitions and consistent approaches to validation.

AI classification, validation and regulatory alignment

Successful use of AI relies on the ability for the experts using it to know it is validated, trusted and accurate – in the lab, once launched and at scale. Trust is not established through technical performance alone. It requires clearly defined use cases, shared terminology and validation and governance applied consistently across organizations and over time. 

This was a central theme at DIA 2026, where the global DIA Artificial Intelligence Consortium brought together regulators, trial sponsors, clinical research organizations, academia and technology providers to develop practical, shared approaches for validating and scaling trustworthy AI.

From principles to practical validation frameworks

A key focus has been creating agreed-upon, practical frameworks to classify AI use cases and apply risk-proportionate validation designed to support real-world deployment and regulatory expectations.

A core foundation of this work is the role of shared terminology in scaling trusted AI. The Consortium has developed a unified glossary of AI terms for life sciences. Inconsistent language and definitions – common across multidisciplinary stakeholders – creates misalignment even if underlying approaches are similar. A common vocabulary ensures clearer communication, reduces ambiguity and supports consistent application of validation and governance frameworks.

Building on shared terminology, the Consortium advanced a structured approach to use-case characterization, recognizing that validation must be anchored to the problem AI is designed to solve. A standardized template to define AI use cases was developed, explicitly specifying purpose, context, risks of incorrect or incomplete outputs, human oversight, performance metrics and desired outcomes and downstream impact. This ensures AI is evaluated in context, with validation requirements aligned to how the model is applied (e.g., lower-risk operational uses or informing regulatory decisions). 

These two foundations, shared terminology and structured use-case definition, enable a structured, risk-based approach to validation. The Consortium’s third deliverable, a five-stage validation framework, recognizes use case characterization as a prerequisite to any validation activity, and creates a practical lifecycle-based approach to validation, where:

• Depth and type of evidence are aligned to the intended use and risk.
• Validation is not a single point-in-time activity but a continuous process spanning development, deployment and ongoing monitoring. 
• Experts and workflow owners contribute at different stages. 

This lifecycle perspective is critical. AI systems evolve over time according to changes in data, workflows and context of use. Sustained trust therefore depends not only on initial validation but on continuous human oversight and reassessment.

The Consortium panel highlighted AI-assisted literature review to illustrate how AI governance can be operationalized with lifecycle-based validation and governance applied in practice. AI can accelerate systematic reviews by screening large volumes of literature and extracting relevant data. The validation framework was applied to this use case to illustrate clearly defined intended use, fit-for-purpose validation and a structure for monitoring performance over time. 

Beyond technical capability, valid use of AI depends on sharing structured approaches that can be  consistently applied across organizations. The Consortium’s multi-stakeholder collaborative model is particularly valuable because it enables early alignment between regulators and industry on terminology, use case definitions and validation expectations.

Looking ahead, the focus is shifting from framework development to real-world settings where challenges related to workflow integration, bias, model performance over time, and user adoption become more apparent. Scaling AI will depend on the ability to operationalize shared terminology, structured use case characterization, and lifecycle-based, risk-aligned validation frameworks in a way that is transparent, consistent and trusted.

The convergence point: designing, measuring and trusting AI-driven trials

What emerges across these perspectives is not a set of isolated innovations, but an interconnected system spanning the full clinical development lifecycle. Simulation improves decisions before trials begin, structured patient data supports reliable evidence generation, and governance frameworks enable trusted, transparent AI deployment.

As the industry moves from experimentation to mature AI adoption, the real transformation lies in how these elements come together. AI’s value will be determined by how well it is governed and embedded into the workflows that generate, evaluate and interpret evidence. For R&D, the focus is less on adoption and more on operationalizing AI to deliver consistent, trusted impact.

Julien is an epidemiologist focused on advancing the use of real-world data, evidence generation and AI-driven approaches to support study design, validation and decision-making. Her work emphasizes rare disease research, regulatory-relevant evidence and the responsible application of advanced analytics and AI in healthcare. She is a member of DIA’s global Artificial Intelligence Consortium focused on advancing responsible and scalable AI in drug development. 

Christina leads scientific and technical innovation in the use of advanced analytics and AI to improve clinical evidence, patient outcomes and healthcare decision-making. Trained as an epidemiologist and computer engineer, she brings deep expertise in real-world data, evidence generation and AI-driven insights. She co-chairs DIA’s global Artificial Intelligence Consortium focused on advancing responsible and scalable AI in drug development.