ICH E6(R3) does not name artificial intelligence specifically, but its principles apply directly to any AI tool used in a clinical trial. The guideline asks that computerized systems be fit for purpose, validated in proportion to risk, and governed so that data integrity and participant safety are protected. For AI, that standard is both flexible and demanding.
The revised guideline, adopted by ICH in January 2025 and now being implemented by regulators, was written to be technology-agnostic. That was deliberate. Rather than regulating specific tools, it sets expectations that any tool must meet.
What does "fit for purpose" mean for AI?
A fit-for-purpose system does what it is intended to do, reliably, within a defined context. For AI, that means three things.
First, the intended use must be clearly defined. "Supports monitoring" is too vague. "Ranks sites by predicted data quality risk to inform the frequency of on-site visits" is specific enough to validate.
Second, validation should be proportionate to the consequences of error. A tool that suggests TMF document names carries less risk than one that triages safety reports. The validation effort should reflect that difference.
Third, performance must be maintained over time. Machine learning models can drift as data changes. A model validated at study start may behave differently six months later. Periodic performance review is part of fitness for purpose, not an optional extra.
Risk-based thinking applies to the tools themselves
E6(R3) places strong emphasis on identifying critical-to-quality factors and focusing effort where risk is highest. The same logic applies to technology. Sponsors should maintain an inventory of computerized systems used on each trial, including AI-enabled vendor tools, and assess each for its potential impact on participant safety and data reliability.
This connects with FDA's January 2025 draft guidance on using AI to support regulatory decision-making, which proposes a risk-based credibility assessment built around a defined context of use. The two documents point in the same direction: define the question, assess the risk, match the evidence to the risk.
Sponsor responsibility does not transfer
E6(R3) reinforces that sponsors retain responsibility for activities delegated to service providers. If your CRO or technology vendor uses AI, you need sufficient understanding and documentation to show the system was appropriate. You do not need to understand the model's mathematics. You do need to understand its purpose, its validation, its limitations and its controls.
Practical steps for sponsors
Add AI-enabled tools to your system inventory for each study.
Request vendor validation summaries tied to intended use.
Define oversight activities in proportion to each tool's risk.
Ensure audit trails capture AI-assisted actions and human review.
Include AI tools in your inspection readiness preparation.
E6(R3) gives sponsors room to adopt innovation. It also expects them to own it.
