Answers · By Hugh Donatello, Atlas Bio · Updated 2026-10-29

How Does FDA's AI Credibility Assessment Framework Work?

How does FDA's AI credibility assessment framework work? FDA's January 2025 draft guidance on AI to support regulatory decision-making for drug and biological products proposes a risk-based, seven-step framework: define the question of interest, define the AI model's context of use, assess model risk, plan how to establish credibility, execute the plan, document the results and deviations, and decide whether the model is adequate for that context of use. Credibility is judged for one context of use at a time, not for a model in general. The guidance is a draft, not for implementation.

What the guidance is and its status

The document is Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products, issued in January 2025 under docket FDA-2024-D-4689 and announced in the Federal Register on 7 January 2025. CDER prepared it in collaboration with CBER, CDRH, CVM, the Oncology Center of Excellence, the Office of Combination Products and the Office of Inspections and Investigations. FDA's guidance page labels it Draft Level 1 Guidance, not for implementation. The comment period requested in the notice closed on 7 April 2025, and FDA's page says comments can still be submitted at any time.

As of 19 October 2026, FDA's guidance page still lists it as a draft, and we found no final version on FDA's site. FDA guidance is nonbinding even when final, and a draft can change before it is finalised, so check the status on FDA's page before relying on any detail here.

What it covers and what it does not

The guidance covers AI used to produce information or data that supports a regulatory decision about the safety, effectiveness or quality of a drug or biological product, across nonclinical, clinical, post-marketing and manufacturing work. It does not address AI used in drug discovery, or AI used for operational efficiencies, such as drafting a submission, that do not affect patient safety, drug quality or the reliability of study results. Whether an AI model is itself a medical device is outside its scope; for devices, FDA points sponsors to device-specific guidance.

The central idea is context of use: the specific role and scope of the AI model used to address a question of interest. FDA's announcement puts it as how an AI model is used to address a certain question of interest. Credibility, in the guidance's sense, is trust in the model's performance for a particular context of use, built from credibility evidence. A model that is credible for one context of use is not automatically credible for another.

The seven steps

The step names below follow the guidance; the descriptions are our summary. The first three steps scope the problem and set the risk, step 4 plans the credibility work, steps 5 and 6 carry it out and record it, and step 7 is the decision.

StepWhat it involves
1. Define the question of interestState the specific question, decision or concern the AI model will help address.
2. Define the context of use for the AI modelDescribe what will be modelled and how the model's output will be used, including whether other information, such as animal or clinical studies, will be used alongside it.
3. Assess the AI model riskCombine model influence, the contribution of the model's evidence relative to other evidence, with decision consequence, the significance of an adverse outcome from an incorrect decision.
4. Develop a plan to establish the credibility of AI model output within the context of useDescribe the model, how it was developed, including the data used, and how it will be evaluated, including limitations and potential biases. The guidance recommends tailoring the plan to the context of use and scaling it to model risk.
5. Execute the planCarry out the planned credibility activities. The guidance says discussing the plan with FDA before execution may help set expectations and identify challenges.
6. Document the results of the credibility assessment plan and discuss deviations from the planCompile the results, covering steps 1 to 4 and any deviations from the plan, in a credibility assessment report. The guidance says the sponsor should discuss with FDA whether, when and where to submit it. It may go in a regulatory submission or meeting package, or be held and made available to FDA on request, for example during an inspection.
7. Determine the adequacy of the AI model for the context of useDecide whether credibility has been sufficiently established for the model risk in this context of use.

How model risk scales the work

Model risk is the possibility that the model's output leads to an incorrect decision with an adverse outcome, not a risk intrinsic to the model. It drives how much evidence the plan needs: higher model influence and higher decision consequence call for more rigorous credibility activities and more detailed documentation. Lower influence or lower consequence reduces model risk, but one does not cancel the other. In the guidance's manufacturing example, an AI visual check of vial fill volume is backed by independent fill-volume testing on a sample of each batch, so model influence is low; but fill volume is a critical quality attribute, so decision consequence is high, and model risk comes out as medium.

If the sponsor or FDA finds that credibility is not sufficiently established for the model risk, the guidance lists five possible outcomes: reduce the model's influence by adding other types of evidence; increase the rigour of the credibility assessment or add more development data; put controls in place to mitigate risk; change the modelling approach; or treat the model's output as inadequate for that context of use, so the context of use is rejected or revised.

After the assessment: life cycle maintenance and early engagement

AI use is usually assessed on locked data at a point in time. Where a model is used over the product life cycle, as in manufacturing, the guidance calls life cycle maintenance critical: planned activities to monitor the model's performance and keep it suitable for its context of use, because performance can change over time or across deployment environments. Depending on the change, parts of the plan may need to be re-run, including retraining and retesting. For manufacturing, the guidance says changes to the model, or manufacturing changes that may affect it, should be evaluated through the manufacturer's change management system within its pharmaceutical quality system, and changes that affect model performance should be reported to FDA under the applicable regulatory requirements.

FDA strongly encourages sponsors to engage early, and the framework envisions FDA feedback on the model risk assessment and the adequacy of the credibility plan. The guidance points to existing meeting and programme routes, chosen according to the intended use. It also notes that for some uses, such as postmarketing pharmacovigilance, sponsors may complete all the steps without early engagement and keep the documentation available to FDA on request.

Using the framework in practice

The first three steps set the target for everything after them; the guidance says a plan discussed with FDA early should at minimum include the question of interest, context of use and model risk, plus the proposed credibility assessment activities based on them, which at that stage can be high-level. Write the question and context of use down precisely enough that a reviewer could tell whether a given output is in scope, and record the model influence and decision consequence reasoning that sets the risk level. This page summarises a draft guidance; it is not legal or regulatory advice, and decisions on regulatory strategy belong to your regulatory function.

The draft guidance recommends that data used to develop an AI model be fit for use: relevant, and reliable, meaning accurate, complete and traceable. An operational record of AI use supports traceability but does not by itself establish model credibility. Atlas Bio's Secure AI for Life Sciences starts with an AI risk assessment of current AI use and data exposure. Status, October 2026. In use in Atlas Bio's own agent platform today: a separate verification step that checks agent work, and a hold that stops a task until a person decides. Built and tested on synthetic data, not yet used with a client: the AI risk assessment tooling (AI discovery, data-flow and vendor-terms mapping, and a NIST AI RMF gap report) and a record of what data an AI used, what it produced and who reviewed it. In development: a customer-controlled AI environment in your own Azure tenant, an AI gateway, agent permissions, evidence linkage and notifications. Each engagement states which controls are built. Decisions on regulatory strategy, credibility evidence and compliance stay with your regulatory and quality functions.

Is FDA's AI credibility guidance final?

No. As of 19 October 2026 FDA's page lists the January 2025 guidance as Draft Level 1 Guidance, not for implementation. Check the page for any update before relying on it.

What does context of use mean in FDA's AI guidance?

The specific role and scope of the AI model used to address a question of interest. Credibility is assessed for that context of use, so a change in how the model is used can mean reassessing it.

How does FDA define AI model risk?

As a combination of model influence, the contribution of the model's evidence relative to other evidence, and decision consequence, the significance of an adverse outcome from an incorrect decision. Model risk is the possibility that the model's output leads to an incorrect decision with an adverse outcome, not a risk intrinsic to the model. Higher risk calls for more rigorous credibility evidence.

Does the guidance apply to AI used in drug discovery?

No. The draft says it does not address AI used in drug discovery, or AI used for operational efficiencies that do not affect patient safety, drug quality or the reliability of study results.

Does the guidance replace 21 CFR Part 11 or apply to medical devices?

It does not replace Part 11. Guidance cannot replace a regulation, and Part 11, which covers electronic records and signatures, still applies to in-scope records; this draft does not discuss Part 11. On devices, the draft is written for drugs and biological products, including the drug part of a combination product, and says its recommendations may also be relevant to devices used with drugs. Whether an AI model is itself a device is outside its scope, and FDA points device sponsors to device-specific guidance.

  1. FDA draft guidance: Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products (January 2025, guidance page)
  2. FDA draft guidance (PDF, January 2025)
  3. Federal Register, 7 January 2025: notice of availability of the draft guidance, docket FDA-2024-D-4689
  4. FDA news release, 6 January 2025: FDA Proposes Framework to Advance Credibility of AI Models Used for Drug and Biological Product Submissions
  5. FDA CDER: Artificial Intelligence for Drug Development

Talk to Atlas Bio.

Sponsors, researchers and clinicians: tell us the decision you are trying to make and we will show what the platform can and cannot answer.

Contact Us →