COO, AG Mednet
The FDA and EMA just got a lot more specific about what they expect when you deploy AI in a clinical trial. And the gap between what regulators now require and what most organizations can actually demonstrate is wider than the industry wants to admit.
Recently, Clinical Leader published a detailed analysis of the evolving regulatory expectations for AI use in clinical trials across both the FDA and EMA jurisdictions. The takeaway isn’t that regulators are blocking AI adoption — it’s that they’re demanding the same rigor for AI-driven decisions that they’ve always demanded for human ones. Validation documentation. Reproducibility evidence. Audit trails that connect a model’s output to the clinical decision it informed.
What regulators are actually asking for
For anyone who has spent time in clinical operations, this shouldn’t be surprising. The regulatory framework has always been built around accountability — knowing who did what, when, and why. The challenge is that most AI implementations in clinical trials today were designed to optimize speed or accuracy, not to produce the documentation trail that a regulatory submission requires.
The real bottleneck isn’t the model
This is where the conversation needs to shift. The bottleneck for AI adoption in clinical trials was never the models themselves. It was never the science. It’s the operational infrastructure — the workflow layer that sits between an AI system’s output and the regulated process it’s supposed to improve. Without that layer, you get impressive demos that can’t survive a regulatory audit.
Even Google is pointing to the infrastructure problem
Google Cloud’s life sciences team acknowledged this explicitly. Their head of life sciences strategy, Shweta Maniar, described the critical gap as “operational integration” — connecting AI capabilities to existing clinical data management systems, submission workflows, and cross-organizational collaboration. Google has the models. They know the integration layer is what’s missing.
Built for this from the start
At AG Mednet, this is the problem we’ve been solving for years with Judi. We built a platform for regulated environments from the ground up — not because we anticipated the current AI wave, but because clinical trials have always required workflow orchestration with accountability built in. The audit trails, the quality controls, the multi-stakeholder governance — those aren’t features we added for AI. They’re the architecture. And that architecture is exactly what regulators are now demanding for any AI system that touches a clinical trial. It’s the same operational argument we’ve made about the FDA’s real-time review pilot — the data has to be governed continuously, not compiled at submission time.
The governance layer is the competitive advantage
The organizations that will deploy AI in clinical trials at scale won’t be the ones with the most sophisticated models. They’ll be the ones with the operational infrastructure to govern those models within a regulated workflow. The governance layer isn’t the boring part — it’s the part that determines whether your AI actually makes it into production.
The regulatory direction is clear. The question is whether your infrastructure is ready for it.
John Paul (JP) Lee, COO, AG Mednet. McKinsey alum and Kellogg MBA, JP drives operational strategy for Judi across regulated clinical trial environments.
