What the FDA and EMA AI Guidance Means for Your Regulatory Team

Regulators have moved from "wait and see" to active direction on AI. Two documents anchor the landscape: the FDA's January 2025 draft guidance on AI for regulatory decision-making, and the EMA's September 2024 reflection paper on AI across the medicinal product lifecycle. Read together, they send one consistent message — and it has practical consequences for how your team adopts AI.

The FDA's risk-based credibility framework

In January 2025 the FDA issued its first dedicated draft guidance, "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products." It applies across the nonclinical, clinical, post-marketing and manufacturing phases whenever an AI model produces information used to support a regulatory decision about safety, efficacy or quality.

Its core is a risk-based credibility approach. In essence, sponsors are expected to:

• Clearly define the regulatory question the model addresses and its context of use;
• Assess model risk — a function of the model's influence on the decision and the consequences of an error;
• Develop and execute a credibility assessment plan proportionate to that risk;
• Document the outcomes, and maintain credibility across the model's lifecycle.

The headline: the more your AI influences a high-stakes decision, the more evidence you need that it is fit for that specific purpose.

The EMA's reflection paper

The EMA's reflection paper, finalised in September 2024, covers AI from drug discovery and clinical development through manufacturing and post-authorisation activities such as pharmacovigilance. It similarly champions a risk-based approach to developing, deploying and monitoring AI/ML tools, alongside strong themes of human oversight, data integrity, transparency, GxP alignment and clear sponsor accountability.

Where they agree — and that's the signal

The specifics differ, but the direction is strikingly consistent across both regulators:

• Risk-based, context-specific. There's no blanket rule — what's required scales with how the model is used and what's at stake.
• Credibility and validation evidence. You must be able to show the model is fit for its intended use.
• Human oversight. AI assists; qualified humans remain accountable for decisions.
• Data integrity and transparency. Provenance, quality and explainability matter.
• Lifecycle thinking. Credibility isn't a one-time check — it's monitored as the model and data evolve.
• The sponsor stays responsible. Outsourcing to a vendor or model doesn't transfer accountability.

For EU operations there's also the EU AI Act layered on top, adding obligations for higher-risk systems. And internationally, regulators are converging toward shared "good AI practice" expectations — so building to this standard now is future-proofing, not over-engineering.

What this means for your regulatory team, in practice

• Inventory your AI. Map where AI is already used or planned across regulatory, quality and PV workflows.
• Classify by risk and context of use. For each use, define the question it answers and how much it influences a regulated decision.
• Build proportionate credibility evidence. Higher-risk uses need stronger validation and documentation; low-risk uses need less.
• Keep humans in the loop — and document it. Define who reviews and approves AI-assisted outputs, and record that oversight.
• Put governance in place. An AI policy, SOPs, accountable owners, change control and clear data-handling rules.
• Monitor over the lifecycle. Track performance, drift and incidents; review periodically.

The bottom line

None of this says "don't use AI." It says: use it deliberately, prove it's credible for its purpose, keep people accountable, and document the lot. Teams that build this scaffolding now will adopt AI faster — and walk into inspections with confidence — while those who bolt it on later will keep stalling.

Primary sources: FDA, Artificial Intelligence for Drug Development (Jan 2025 draft guidance); EMA, Reflection paper on the use of AI in the medicinal product lifecycle (Sept 2024). This article is general information, not legal or regulatory advice — always consult the primary guidance and your own advisors.