Note from the author: I wrote this article as an extension to last week's fun and thought-provoking conversation with Brad Hightower and Denali Rose on the Note to File podcast. Watch the full episode here.

The question every site leader is quietly asking

There’s a question a lot of research sites are quietly asking right now: what would a sponsor think if they found out we were using AI on their trial data at all?

It’s a fair worry. A site’s reputation rests on the integrity of its data, and “AI” is a word that makes people in a high-stakes field uneasy. But here’s the short answer, and the rest of this post is the evidence for it: using a well-built AI tool doesn’t change who is accountable for your data, what counts as the record, or what your sponsor is owed. A trained person still reviews every value and signs their name to it. The tool simply makes the work in between faster and less error-prone.

This is exactly what the FDA and modern Good Clinical Practice already ask for. And, as of this year, it’s what the FDA has said in writing when it has looked at AI use directly.

You already stand behind every data point

Start with the fact that resolves most of the worry. Under 21 CFR 312.62 and the commitments your investigator signs on Form FDA 1572, the site already stands behind every data point in the study. That has been true for every tool your site has ever used: your EHR, your eSource, your spreadsheet.

A tool like Gleam keeps that responsibility with you, exactly where it has always sat, rather than shifting it onto an algorithm or a vendor. So when you tell a sponsor you use AI-assisted data entry, nothing about who answers for the data has moved. It sits where it always has, with you, and that is what makes the conversation an easy one.

Two questions decide whether any tool is acceptable

Strip away the word “AI” and regulators really ask only two things of any technology that touches trial data: "Is it fit for purpose?" And, "Is a qualified human accountable for the result?" Hold any tool to those two tests.

On the first, ICH GCP E6(R3) — the modern GCP standard — asks that computerized systems be “fit for purpose” and that their quality be built into the design. 21 CFR Part 11 and the FDA’s 2024 final guidance on electronic systems ask the same thing in risk-based terms: records that are attributable, accurate, complete, and secure, with an audit trail and validation appropriate to the risk. Both standards are deliberately technology-neutral. Neither names AI, just as neither names Excel.

On the second, every output is reviewed by a person before it becomes the record. That is the test the FDA has been most explicit about, as the rest of this post shows.

What a well-built tool actually does

The picture people fear is an algorithm quietly inventing data and pushing it into the EDC while no one watches. A well-built tool works nothing like that.

Gleam is a 21 CRF Part 11-validated data-entry assistant focused on quality-control. It reads the source document you point it at, drafts the entry, and shows its work: every value is highlighted and linked back to exactly where it came from, so a coordinator can confirm it against source in one click. A trained human reviews it and saves it. The system never submits, approves, or finalizes anything on its own. Learn, Fetch, Fill, Trace — with a person at the wheel the whole way.

In other words, it passes both tests: built for one narrow, validated job, with a qualified human accountable for every result. It belongs to the same family of clerical support as the calculators and auto-population tools your site already uses every day, just with a cleaner record of where each value came from.

The “FDA is cracking down on AI” headlines are about something else

The headlines are real, but they are about a different category of tool. The FDA’s January 2025 draft guidance on AI covers models that produce information used to support a regulatory decision about a drug’s safety, effectiveness, or quality — for example, a model used to justify an endpoint or a dose in a submission. That is a high bar, and it should be. Moving a blood-pressure reading from a source document into the EDC, with a human verifying it and a trail back to its origin, is not that.

And when the FDA has looked at AI use directly in an enforcement setting, the line it drew was exactly the human-in-the-loop one. In an April 2026 warning letter to a drug manufacturer, the company was cited because it had relied on AI to generate compliance documents without a qualified person reviewing the output; the use of AI itself was never the issue. The FDA’s instruction was explicit: any output or recommendations from an AI agent “must be reviewed and cleared by an authorized human representative.” That case concerned drug manufacturing under CGMP rather than clinical data entry, so it doesn’t bind a trial site directly. But it is the clearest signal yet of how the agency thinks about AI accountability, and the principle is identical.

For a site, that warning letter should be reassuring. The failure it punished was the absence of human review — exactly the safeguard a tool like Gleam is built around, where a person reviews and clears every value before it stands.

The data actually comes out cleaner

Step back from the word “AI” and ask what a sponsor’s data team actually wants: accurate data, a complete audit trail, fewer queries, a faster path to database lock.

Manual transcription works against all four. Re-keying numbers by hand is one of the oldest and most reliable sources of error and query volume in our industry. A traceable, human-reviewed assistant attacks that directly: fewer transcription slips, enforced consistency, and a complete, time-stamped record of who entered what and where it came from. It’s also the direction the rules are already moving — E6(R3) and the FDA’s guidance on electronic source data discourage unnecessary transcription between the source record and where the data lands. A tool that pulls from your source, fills the EDC, and keeps a clean line back to the original is doing exactly what that guidance points toward. Done well, AI-assisted entry reinforces ALCOA+ data-integrity principles. And when an inspector asks where a value came from, a complete trace answers in one click, making source verification faster.

How to tell your sponsor

You don’t need to bury the fact that you use an AI-assisted tool, and you don’t need to oversell it. The honest, accurate version is also the reassuring one:

• The site remains accountable for every data point, exactly as it always has been (21 CFR 312.62; Form FDA 1572).

• The tool is fit for purpose and used under human review. A qualified person verifies and saves every value, consistent with ICH GCP E6(R3) and 21 CFR Part 11.

• It falls outside the FDA’s guidance on AI in regulatory decision-making: it generates no predictions, interpretations, or decisions used to support a regulatory determination, and only relocates existing source data with a trail back to the original.

• Gleam is validated to 21 CFR Part 11, and the validation package is available on request to support the sponsor’s own oversight. Compliance here is a shared responsibility under FDA guidance and ICH GCP. The vendor supplies validated software and documentation, while the site maintains access, training, and signature controls.

One practical note: some sponsors set their own contractual terms about using or disclosing AI tools. Your study agreement is the place to confirm what applies, and an early word with your sponsor is the easiest way to stay ahead of it.

That’s a position you can put in front of a sponsor, a monitor, or an inspector without flinching. The tool makes your data cleaner and your trail clearer, and it leaves accountability exactly where the regulations already put it: with you.

References & Citations

• 21 CFR 11.10: Controls for closed systems (Electronic Records; Electronic Signatures): “Persons who use closed systems … shall employ procedures and controls designed to ensure the authenticity, integrity, and, when appropriate, the confidentiality of electronic records … [including] (a) Validation of systems to ensure accuracy, reliability, consistent intended performance, and the ability to discern invalid or altered records.”

• 21 CFR 312.62(b): Investigator recordkeeping and record retention: “An investigator is required to prepare and maintain adequate and accurate case histories that record all observations and other data pertinent to the investigation on each individual administered the investigational drug or employed as a control in the investigation …”

• FDA: “Electronic Systems, Electronic Records, and Electronic Signatures in Clinical Investigations: Questions and Answers,” final guidance, October 2024: “… we recommend that regulated entities use a risk-based approach for validating the electronic systems they deploy.”

• ICH E6(R3) Good Clinical Practice — FDA final guidance, September 2025 (Principle 9.3): “Computerised systems used in clinical trials should be fit for purpose (e.g., through risk-based validation, if appropriate), and factors critical to their quality should be addressed in their design or adaptation for clinical trial purposes to ensure the integrity of relevant trial data.”

• FDA: “Electronic Source Data in Clinical Investigations,” guidance, September 2013: “Capturing source data electronically and transmitting it to the eCRF should: … Eliminate unnecessary duplication of data … Reduce the possibility for transcription errors … Eliminate transcription of source data prior to entry into an eCRF …”

• FDA: “Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products,” draft guidance, January 2025: the guidance addresses “the use of AI models … where the specific use of the AI model is to produce information or data to support regulatory decision-making regarding safety, effectiveness, or quality for drugs.”

• FDA: Warning Letter to Purolea Cosmetics Lab (MARCS-CMS 722591), April 2, 2026: “… any output or recommendations from an AI agent must be reviewed and cleared by an authorized human representative of your firm’s QU …” (fda.gov)