In April 2026 FDA asked the public how a pilot using AI to improve early-phase trials should be designed. It is a request for information, not a rule and not yet a programme you can join. Here is what it said, what we could not confirm, and the data habits worth building now.
Free sandbox · No credit card · 21 CFR Part 11 aligned
days
Published 29 April 2026
Extension notice late May
Check the docket
The pilot in six points
The notice
FDA's notice, AI-Enabled Optimization of Early-Phase Clinical Trials Pilot Program; Request for Information, appeared in the Federal Register on 29 April 2026 and is cited as 91 FR 23100. It asks for input on a proposed pilot to assess how AI-enabled technologies can improve efficiency, speed and quality of decision-making in early-phase clinical trials. The areas it highlights are trial efficiency, safety monitoring, dose selection decisions and more informed early go/no-go decisions. FDA said the pilot would be guided by principles aligned with the NIST AI Risk Management Framework.
The comment period was extended to 29 June 2026 by a notice dated late May. Some secondary sources still list 29 May as the deadline, which was the original date as reported, so use the later date and the Federal Register notices themselves as the authority. Trade bodies and research organisations filed comments, including an association of contract research organisations on 24 June 2026.
Be clear about what this is not. It is not guidance and it does not say how FDA will review AI-derived evidence in an IND. Our companion post, FDA's AI guidance for clinical trials, covers the January 2025 draft guidance on AI used to support regulatory decision-making, which is the document that sets out FDA's credibility framework. The pilot sits next to it and would, if it goes ahead, give FDA practical experience with early-phase use cases.
The questions
Summarised from secondary descriptions of the notice. Check the Federal Register text for the precise questions.
| Area | Topics reported | Why a small sponsor might care |
|---|---|---|
| Pilot design and implementation | Scope and focus, how participants would be selected, collaboration models, operational structure, timeline and milestones, knowledge sharing | Shows whether a pilot would be open to small sponsors or shaped around large programmes |
| Evaluation and success criteria | Metrics and criteria for judging whether AI actually improved decisions | Hints at what evidence FDA would want: faster decisions are not enough on their own |
| Use cases | Safety monitoring, dose selection decisions, early go/no-go decisions, overall trial efficiency | These are the decisions an early-phase biotech makes with very few participants and very little time |
| Governance | Trustworthiness concepts from the NIST AI RMF, and how current GCP and governance structures cope with AI inside trial conduct | Points to documentation of what the AI does, who reviews it and how it is controlled |
One secondary summary also described the RFI as asking about data pipelines, interoperability, validation checks and audit trails. We could not confirm that wording against the notice.
Build a first-in-human style study in the free sandbox: every feature, no credit card, and you pay only when you go live with real participants.
Our reading
Early-phase trials turn on a handful of fast decisions: whether a dose cohort was tolerated, whether to escalate, whether the signal justifies the next study. Those decisions are made by a small team on data from a few dozen participants, often under time pressure and often with data that arrives late or in several places. That is where an AI-assisted tool could plausibly help, and it is also where an unexplained tool could do the most damage.
That is our interpretation, not FDA's. The RFI does not say that small sponsors are a target group. What it does show is a regulator asking, before any rule exists, how to judge whether AI helped. The likely answer will be evidence about data quality, human oversight and traceability, which are things a small sponsor can influence long before any pilot opens.
What to do now
None of this is required by the RFI. It is sensible practice that also happens to be what any future AI evaluation would ask for.
List the decisions your protocol makes at each cohort: dose escalation, stopping rules, safety reviews. Note what data each decision uses and when it needs to be ready.
Enter data at the site as it happens, review queries daily during dosing windows and make sure safety events are in the same system as the clinical data, not in a separate spreadsheet.
If you use AI for protocol drafting, form building, data review or anything else, record the tool, the purpose, who reviews the output and what happens when it is wrong. Our GxP checklist for AI in data capture gives a structure.
Treat outputs as drafts. In Capture, for example, the AI study builder drafts visits and forms from an uploaded protocol and nothing is saved without human review.
Confirm the audit trail records user, timestamp, old value, new value and reason for change, and that you can export it. See audit trail software for clinical trials.
Set a reminder to check docket FDA-2026-N-4390 and the FDA early-phase pages. If a pilot opens, you want to read the criteria before you decide.
Worked example
Imagine a six-participant dose cohort in a first-in-human study. The sponsor wants the data needed for a cohort review to be reviewed and query-free before the meeting. The panel below is a demo of the kind of status view a data manager checks, with invented numbers.
Participants dosed
6 / 6
Open queries
3
Overdue entries
0
Safety events awaiting review
1
2 pending
1 pending
Where Capture fits
Capture is a self-serve, all-in-one platform with EDC, eCRF building, ePRO, eConsent, safety reporting and monitoring workflows in one system with one audit trail, used from Phase 1 to Phase 3. For early-phase teams that means dosing, safety and participant-reported data can sit together. See EDC for first-in-human studies and clinical trial software for early-phase trials.
Capture has an AI study builder that drafts a visit schedule and forms from an uploaded protocol, with nothing saved until a person reviews it. It applies 21 CFR Part 11-aligned controls and a field-level audit trail. We do not claim any role in FDA's pilot, and nothing on this page says Capture is, or will be, a pilot participant or that FDA has endorsed any product. For wider context on AI that acts on trial data, read our AI agents in EDC reality check, and for the EU view the EU AI Act and clinical trial software.
It is a proposed pilot that FDA described in a request for information published in the Federal Register on 29 April 2026 (docket FDA-2026-N-4390). The aim is to assess how AI-enabled technologies can improve the efficiency, speed and quality of decision-making in early-phase trials.
We could not confirm that it is. The notice was a request for information, and we found no published selection criteria or participant list as of 16 October 2026. Check the docket and FDA pages for updates.
The comment period was extended to 29 June 2026. Some secondary sources show 29 May 2026, which was the original reported date. Use the Federal Register notices as the authority.
No. An RFI collects views and creates no requirements. FDA's January 2025 draft guidance on AI to support regulatory decision-making is a separate document and is also not final guidance in the sense of a settled requirement.
Nothing in the notice changes existing regulations. 21 CFR Part 11 and GCP expectations continue to apply to electronic systems, including any that use AI.
Shorten the path from visit to clean dataset, document every AI use with a context of use and a human reviewer, keep a field-level audit trail and watch the docket. These steps help whether or not a pilot opens.
We make no such claim. Capture offers an AI study builder that drafts forms for human review and a Part 11-aligned audit trail, but FDA has not endorsed any product for this pilot.
Keep exploring
FDA's AI guidance for clinical trials
The wider regulatory record on AI.
AI agents in EDC: a reality check
What agents acting on trial data can and cannot do.
GxP checklist for AI in data capture
Controls to document before you use AI.
Clinical trial software for early-phase trials
What early-phase teams need from software.
EDC for first-in-human studies
Dose-cohort data capture.
Audit trail software for clinical trials
What a defensible audit trail records.
Every feature, no credit card, and you pay only when you go live with real participants.