AI Tools in Clinical Trials: FDA 2025 Guidance

Key Takeaways

• FDA’s January 2025 draft AI guidance sets a risk-based credibility framework for models used in regulatory decision support.
• CDER reports experience with hundreds of AI-containing submissions between 2016 and 2023, signaling rising review familiarity.
• Operational efficiency AI is not the same as AI that generates evidence for safety, effectiveness, or quality.
• Sponsors should lock context of use, data lineage, and human oversight before scaling AI into pivotal trials.

What does FDA’s 2025 AI draft guidance actually cover?

The draft guidance titled Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products recommends a credibility assessment approach tied to a defined context of use (COU). Higher-impact uses—such as endpoint adjudication or analyses that affect patient safety conclusions—require stronger evidence that the model is fit for purpose.

Primary source: FDA draft guidance page.

Where are sponsors already using AI in trials?

Common use cases include site selection, enrollment forecasting, protocol deviation detection, imaging reads, and synthesis of real-world data streams. FDA’s CDER AI hub notes the Agency’s exposure to more than 500 submissions with AI components from 2016 to 2023 and points to guiding principles developed with EMA.

See FDA CDER Artificial Intelligence for Drug Development and the joint Guiding Principles of Good AI Practice.

What changes for clinical operations leaders in 2026?

• Map every AI tool to a COU and regulatory decision impact score
• Retain human-in-the-loop controls for safety-critical outputs
• Document training data provenance, drift monitoring, and change control
• Align vendor contracts with auditability of model versions used in submissions

Trial registries remain the ground truth for study design transparency. Teams validating AI-assisted recruitment or endpoint tools should still register protocols on ClinicalTrials.gov when applicable.

What remains unproven

FDA draft guidance does not certify any commercial AI vendor. Claims that AI “guarantees” shorter Phase 3 timelines or automatic approval acceleration are marketing, not regulatory fact. Credibility is context-specific and must be shown for each intended use.

Investment and BD implications

Diligence should ask whether AI outputs will appear in CTD modules, whether COU documentation exists, and whether model risk is scored. Prefer platforms that can freeze versions used for a locked database and reproduce analyses for inspection.

Inspection readiness for AI-enabled trials

When AI influences eligibility, dosing recommendations, or endpoint derivation, inspectors will ask who approved the model version used at database lock. Keep a model inventory linked to SOPs, change-control tickets, and training records for site staff who rely on algorithmic prompts.

For APAC-heavy global trials, ensure translations of AI-assisted worksheets do not alter decision logic. Recreate key analyses from archived model outputs during dry-run inspections so surprises do not appear during a real FDA or EMA review cycle in 2026 or later.

Related NovaPharma coverage

• PMDA approves ENHERTU for HER2-low breast cancer
• India invites global pharma partnerships
• Clinical trials outsourcing market outlook

Primary Sources

1. FDA — Considerations for the Use of AI to Support Regulatory Decision-Making
2. FDA CDER — Artificial Intelligence for Drug Development
3. FDA/EMA — Guiding Principles of Good AI Practice in Drug Development

Frequently Asked Questions