What sponsors need to know today

AI isn’t a future consideration in clinical research—it’s already embedded in how trials are designed, executed, and monitored.

From protocol optimization and patient recruitment to data cleaning, endpoint analysis, and safety monitoring, AI is influencing decisions across the entire clinical trial lifecycle.

In many cases, it’s doing so quietly.

AI tools are being used to:

• Identify eligible patients from electronic health records
• Predict enrollment success and optimize site selection
• Analyze endpoints and detect safety signals
• Automate data processing and submission preparation

These capabilities are powerful. They address long-standing inefficiencies in clinical trials, including recruitment delays, high costs, and data quality issues—challenges that affect the majority of studies today.

But the regulatory expectations haven’t changed.

If anything, they’ve become more stringent.

Scope: where AI shows up in clinical trials

This isn’t about AI as a regulated product.

This is about AI being used within regulated clinical trial activities, including:

• Protocol design and feasibility assessment
• Patient recruitment and eligibility screening
• Trial operations and monitoring
• Data analysis and endpoint evaluation
• Safety surveillance and pharmacovigilance

What makes this particularly important is that AI is not confined to one stage. It influences both upstream decisions (like study design) and downstream outcomes (like safety reporting and regulatory submissions).

That creates a continuous chain of risk.

Where regulators stand today

Regulatory expectations for AI in clinical trials are evolving—but there is already strong alignment between the FDA and EMA on core principles.

Both agencies are clear on one point:

If AI influences patient safety, trial integrity, or regulatory decision-making, it is not just a tool—it is part of the regulated system.

This has several implications:

• AI models must be pre-specified in protocols and statistical analysis plans
• Their context of use must be clearly defined
• Performance must be validated against predefined criteria
• Outputs must be explainable and reproducible

Retrospective justification—adding documentation after the fact—is unlikely to be accepted.

Core regulatory expectations for AI in clinical trials

Across FDA and EMA guidance, a consistent set of expectations is emerging.

1. Risk-based validation

AI is not regulated uniformly—it is regulated based on risk.

• High-risk applications (e.g., patient selection, dosing decisions, endpoint determination) require rigorous validation, detailed documentation, and potentially regulatory engagement
• Lower-risk applications (e.g., drafting documents or internal analytics) require less oversight but still need verification

The FDA’s credibility framework reinforces this by requiring sponsors to define the model’s context of use, assess risk, and demonstrate that it is fit for purpose.

2. Transparency and explainability

Regulators expect AI systems to be understandable—not black boxes.

Sponsors must be able to document:

• How the model was built
• What data it was trained on
• How inputs are processed
• How outputs are generated and interpreted

This is particularly important in clinical trials, where decisions must be scientifically justified and reproducible.

The challenge is that many modern AI systems—especially generative models—do not naturally meet these expectations.

3. Data integrity and representativeness

AI is only as reliable as the data it is trained on.

Regulators expect:

• High-quality, curated datasets
• Representation across relevant patient populations
• Controls to mitigate bias

This is especially important in clinical trials, where lack of diversity or biased data can directly impact trial outcomes and generalizability.

AI has the potential to improve recruitment diversity—but it can also reinforce existing biases if not properly managed.

4. Human-in-the-loop oversight

AI does not replace human accountability.

Regulators consistently emphasize the need for:

• Human review of AI-generated outputs
• Clear ownership of decisions
• Oversight mechanisms at critical decision points

While “human in the loop” is currently the default mitigation strategy, it is not without limitations—particularly as scale increases and oversight fatigue becomes a concern.

5. Lifecycle management and continuous monitoring

AI is not static.

Models can drift, degrade, or change behavior over time. As a result, regulators expect:

• Continuous performance monitoring
• Periodic revalidation
• Change management processes

The FDA specifically recommends mechanisms such as Algorithm Change Protocols (ACPs) to manage updates in self-learning systems.

6. Protocol and documentation requirements

One of the most important—and often overlooked—expectations is documentation.

If AI influences trial outcomes, it must be:

• Clearly described in the protocol and statistical analysis plan
• Defined in terms of inputs, outputs, and intended use
• Supported by validation evidence

The EMA explicitly considers AI models used in analysis or endpoint evaluation to be part of the statistical methodology, not external tools.

AI across the clinical trial lifecycle

The regulatory implications of AI become clearer when viewed across the trial lifecycle.

AI in clinical trials: regulatory expectations and risk

The real risk: AI as a hidden variable

The biggest regulatory risk is not AI itself.

It’s AI being used without being formally recognized as part of the trial.

When AI operates in the background—through vendor tools, automation, or analytics—it can quietly influence:

• Who gets enrolled
• How data is interpreted
• How endpoints are evaluated

This creates what is effectively a hidden variable in the trial.

And regulators are increasingly focused on eliminating that risk.

Where the regulatory challenges are

There are several consistent challenges that sponsors must navigate when implementing AI in clinical trials.

• Misalignment between AI capabilities and regulatory frameworks
  Many current regulations were designed for deterministic systems, not probabilistic AI models. This creates uncertainty in how requirements like auditability and validation should be applied.
• Data access, privacy, and intellectual property constraints
  Clinical trial data is highly sensitive and subject to strict legal protections. Balancing data utility with privacy and IP considerations remains a major barrier.
• Lack of transparency in modern AI systems
  Generative AI models often lack clear data provenance and reproducibility, making them difficult to validate in regulated environments.
• Operational vs. regulated use ambiguity
  The line between operational tools and regulated systems is not always clear. A tool used for efficiency can quickly become one that impacts trial integrity.
• Accountability remains with the sponsor
  Even when AI is used by vendors or CROs, responsibility for compliance does not shift. Sponsors remain accountable for outcomes.

What this means for sponsors

AI is not reducing regulatory burden in clinical trials—it is changing it.

Sponsors must now manage not only:

• Study design and execution
• Data quality and compliance

But also:

• AI model validation
• Data governance
• Documentation and traceability
• Vendor oversight

AI introduces new efficiencies—but also new failure points.

How to be prepared

The most effective approach is to treat AI as part of the clinical and quality system.

That means:

• Clearly defining where AI is used in the trial
• Assessing risk based on context of use
• Validating models before deployment
• Embedding AI into protocol and SAP documentation
• Maintaining audit trails and version control
• Monitoring performance throughout the trial lifecycle

Early engagement with regulators is also critical—particularly for high-impact AI applications such as endpoint evaluation or patient stratification.

Conclusion

AI is fundamentally changing how clinical trials are designed, executed, and analyzed.

Regulators are not limiting its use—but they are raising expectations around how it is implemented.

Across both FDA and EMA frameworks, a consistent message is emerging:

AI must be treated as part of the regulated system—not an external tool.

That means it must be:

• Risk-assessed
• Documented
• Validated
• Continuously monitored

For sponsors, the challenge is not whether to adopt AI—but how to integrate it in a way that remains compliant, transparent, and defensible.

How dicentra can help

At dicentra, we support sponsors in navigating the intersection of clinical strategy, regulatory expectations, and emerging technologies like AI.

We help organizations:

• Assess where AI is impacting clinical trial design and execution
• Identify regulatory risks tied to AI use across the lifecycle
• Align protocols, SAPs, and quality systems with FDA and EMA expectations
• Implement governance frameworks for AI validation and monitoring
• Ensure inspection readiness in an evolving regulatory landscape

AI can significantly improve clinical trial efficiency and outcomes.

But only when it is implemented with the same rigor as any other regulated component of the study.

Contact dicentra today for support with AI in clinical trials.