Regulayer / The 143 laws / Pharma & life sciences
FDA · AI credibility assessment (draft)
Risk-based credibility for AI supporting regulatory decisions. Draft guidance, not final.
Instrument: Draft guidance, nonbindingApplies to: Drug and biological product sponsors (US).
What the signed record shows
The context an AI model was used in, and the human decisions taken on its output, recorded as they were made.
The proof is a signed, tamper-evident record. Anyone can check it, free, without an account, and nothing has to leave the building to make it. Evidence, not a promise.
Citation: Considerations for the Use of AI To Support Regulatory Decision-Making for Drug and Biological Products, draft guidance, January 2025
What to do about it
Keep the evidence this asks for, as the work happens.
A consequential AI-proposed action is checked against the human authority in force at that moment, and the decision leaves a signed record that a third party can verify independently, offline, without Regulayer. It runs inside your own environment and nothing has to leave it.
Life sciences: how the control worksOr verify a real record, free →
Part of the Regulayer proof catalogue: 143 laws and standards, one sealed engine. This page is a product description, not legal advice.
What the draft guidance sets out, and what it leaves out
FDA CDER, Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products. Issued as draft guidance in January 2025 and informed by more than 500 submissions containing AI components. As drafted it sets out:
- A seven-step credibility assessment
- Credibility is established through a risk-based framework rather than a fixed technical standard, with the depth of assessment keyed to how much the decision relies on the model.
- Context of use, established first
- The question is not whether a model is good in general but what it is being used for and how much weight the decision places on it. Everything else follows from that definition.
- Scope: nonclinical, clinical and manufacturing
- The draft addresses AI used to support regulatory decisions across those phases.
- What is outside the scope of this document
- The draft expressly does not address AI used in drug discovery, and does not address AI used for operational efficiency. The boundary is stated here rather than left out, and it is a boundary on this document only.
Status: draft, and narrow. This guidance is not final. Its exclusion of drug discovery means that this draft makes no recommendations for discovery AI. It does not mean no obligation of any kind applies there: other requirements can still reach that work, and the EMA reflection paper of September 2024 addresses AI in drug discovery directly. Where AI is used in discovery, the reason to record who decided what, and under what authority, is control and provenance rather than a requirement in this particular document.
Built against the primary text: FDA CDER draft guidance, January 2025. Verified 17 August 2026.
