Draft guidance

AI credibility depends on the question being asked

AI & computationRegulatory science

US Food and Drug Administration · Jan 2025

What the source reports

FDA’s January 2025 draft describes a risk-based approach to evaluating AI models used to support regulatory decisions about drugs and biological products.

Why it belongs here

AI can suggest directions as well as accelerate work. Its evidence must still be credible for the intended use.

WHAT THIS DOES NOT ESTABLISH

Draft, nonbinding recommendations. A technically impressive prediction is not equivalent to demonstrated clinical utility.

Before you use it

Read the original source, check its population or model and intended use, and review uncertainty, disclosures and any subsequent updates. This note is a guide to reading, not a substitute for the paper or guidance.

Open the original source

Keep exploring.

Draft guidance

A validation framework for human-relevant methods

FDA’s March 2026 draft guidance sets out general scientific considerations for validating New Approach Methodologies in drug development and regulatory submissions.

Human models & NAMsRegulatory science
Final guidance

How model-informed evidence should support a decision

ICH M15 addresses planning, evaluation and documentation of model-informed drug-development evidence. FDA’s final guidance is dated June 2026.

AI & computationRegulatory science
Regulatory explainer

A biomarker needs a clearly defined job

FDA distinguishes biomarker categories and explains qualification within a stated context of use. A biomarker can be a molecular, tissue, imaging or physiological measurement.

Proteomics & biomarkersRegulatory science