Talk of an FDA "AI rulebook" for clinical trials has been loud this year. We went to fda.gov and the Federal Register to see what is actually on the record. Short version: the core guidance is still a draft, a lot has happened around it, and there is plenty sponsors can do now.
Free sandbox · No credit card · 21 CFR Part 11 aligned
| Date | Status | |
|---|---|---|
| AI credibility guidance | Jan 2025 | Draft |
| Elsa, FDA internal tool | Jun 2025 | In use at FDA |
| Agentic AI for FDA staff | Dec 2025 | In use at FDA |
| FDA-EMA Good AI Practice | Jan 2026 | Principles |
| Early-phase AI pilot RFI | Apr 2026 | Comments closed |
The short version
The core document
On 6 January 2025 FDA announced Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products, and the Federal Register notice followed on 7 January (docket FDA-2024-D-4689). CDER led it with CBER, CDRH, CVM, the Oncology Center of Excellence, the Office of Combination Products and the Office of Inspections and Investigations. FDA asked for comments by 7 April 2025 so they could inform the final version. In October 2026, the guidance page and the PDF still mark it as a draft, not for implementation, and we found no Federal Register notice announcing a final version.
The scope is specific. The draft covers AI used to produce information or data that supports regulatory decisions about a drug's safety, effectiveness or quality, across nonclinical, clinical, postmarketing and manufacturing work. It does not cover AI used in drug discovery. It also leaves out AI used for operational efficiencies, such as internal workflows, resource allocation or drafting a submission, but only where that use does not affect patient safety, drug quality or the reliability of results from a nonclinical or clinical study. That condition is the part sponsors most often skip.
The draft sets out a risk-based credibility assessment framework:
The draft illustrates step 3 with a trial in which an AI model alone decides which participants can skip 24-hour inpatient monitoring after dosing. Because the model is the sole decision-maker and a wrong call could leave someone with a life-threatening reaction outside hospital, both influence and consequence are high, so model risk is high. The more risk, the more evidence FDA expects. The draft also stresses life cycle maintenance, since model performance can drift as data changes, and it repeatedly encourages early engagement with FDA.
What came after
FDA's own AI use. On 2 June 2025 FDA launched Elsa, a generative AI tool for its staff, which the agency said was already helping with clinical protocol reviews and adverse event summaries. On 1 December 2025 it added agentic AI capabilities for all employees, optional to use and with human oversight built in. In both announcements FDA said the models do not train on data submitted by regulated industry. These tools change how FDA reviews your documents, not what you must submit.
Principles with EMA. In January 2026 FDA and EMA published Guiding Principles of Good AI Practice in Drug Development, ten principles covering the drug product life cycle: human-centric design, a risk-based approach, adherence to standards including GxP, a clear context of use, multidisciplinary expertise, data governance and documentation, model design and development practice, risk-based performance assessment, life cycle management, and clear information for users and patients. They are principles, not guidance with requirements, but they show the EU and US sharing a vocabulary.
An early-phase pilot, still at the idea stage. On 29 April 2026 FDA published a request for information on an AI-Enabled Optimization of Early-Phase Clinical Trials Pilot Program (docket FDA-2026-N-4390). It asked how AI could improve safety monitoring, dose selection and early go/no-go decisions, guided by the NIST AI Risk Management Framework. The comment period was extended to 29 June 2026. As of 8 October 2026 we found no further Federal Register notice on the pilot.
Status check
| Document | Date | What it is | Binding on sponsors? |
|---|---|---|---|
| Considerations for the Use of AI to Support Regulatory Decision-Making (FDA-2024-D-4689) | January 2025 | Draft guidance | No. Draft guidance is non-binding and marked not for implementation |
| Elsa and agentic AI announcements | June and December 2025 | FDA's internal tools | No. They describe how FDA works |
| Guiding Principles of Good AI Practice in Drug Development (FDA and EMA) | January 2026 | Joint principles | No. They are high-level principles |
| RFI on the AI-enabled early-phase trials pilot (FDA-2026-N-4390) | April 2026, comments to June 2026 | Request for information | No. It asks for input |
| 21 CFR Part 11, 21 CFR 312 and other existing rules | Various | Regulations | Yes. They apply to AI-enabled systems like any other |
The last row is the one people forget. Existing rules on electronic records, IND studies and GCP already cover systems that use AI.
Upload a protocol to the free sandbox and review every suggestion before it is saved. No credit card, and you pay only when you go live.
What to do
A draft guidance is not binding, but it is the clearest statement of what FDA reviewers will look for. Sponsors who wait for the final version will be building their evidence late. Five things are worth doing now.
Data capture tools
A common use of AI in EDC and eCOA tools is study build: drafting forms, schedules or edit checks. Used with human review and normal user acceptance testing, that kind of tool is a good candidate for the draft's operational-efficiency exclusion, because the reviewed and tested build, not the model, determines what data is collected. The condition still applies. If AI output went into the study unreviewed, or started acting on participant data, the analysis would change. Our GxP checklist for AI in clinical data capture works through this use by use.
Capture's AI study builder is built that way. It reads a protocol and drafts the visit schedule and forms, nothing is saved without human review, and it only works on draft forms. Reviewed forms then go through the usual draft-to-approved lifecycle before anything reaches a participant, and the data collected afterwards carries a field-level audit trail under 21 CFR Part 11-aligned controls.
What to watch
Three things would change this picture: a final version of the January 2025 guidance, which would appear as a Federal Register notice of availability; a pilot announcement following the early-phase RFI; and EU documents that move in parallel, such as the draft EU GMP Annex 22 on AI and the EU AI Act timetable covered in our EU AI Act guide for clinical trial software. We will update this post when any of them lands.
Not as of 8 October 2026. Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products, issued in January 2025, is still marked draft on FDA's site, and we found no Federal Register notice of a final version.
We could not find one. A search of FDA's guidance pages and the Federal Register shows no new AI guidance or rule for trials that month. The documents that exist are the January 2025 draft, the January 2026 FDA-EMA principles and the April 2026 pilot RFI.
A seven-step, risk-based process in the January 2025 draft: define the question of interest and context of use, assess model risk from model influence and decision consequence, then plan, execute, document and judge the credibility of the AI model's output for that use.
Often not, because it excludes operational uses that do not affect patient safety, drug quality or the reliability of study results. That depends on human review and testing of what the AI produces. If you are unsure, FDA encourages early engagement.
No. They are ten high-level principles published in January 2026 to guide good practice and future guidance. Existing rules such as 21 CFR Part 11 still apply to AI-enabled systems.
Keep exploring
AI in clinical data capture: GxP checklist
Ten checks before an AI feature touches your study.
AI study builder
Protocol to draft study, with human review.
EU AI Act for clinical trial software
The EU side of the picture.
21 CFR Part 11 compliance checklist
The rules that already apply.
FDA digital health technologies guidance
FDA on remote data capture tools.
Free sandbox with every feature. No credit card, and you pay only when you go live.