FDA's First AI Warning Letter: The Model Wasn't the Problem

In April 2026 FDA sent a warning letter to PurOlea Cosmetics Lab that became known as the first to cite a firm's use of AI. Most of the letter is about something else: insects in the production area, finished drugs released without microbiological testing, and components accepted on supplier paperwork alone.

Read the letter

PurOlea was a small operation in Livonia, Michigan, making homeopathic drug products, among them "Dermveda Extra Strength Shingles Relief" and an "Ultra Genital Herpes Relief" product. The disease claims made both unapproved new drugs.

FDA's investigator observed insects, filth, leaves and clutter in several areas of the facility, and a docking bay door that left manufacturing open to the outside. The firm released finished drug products without testing them for microbiological attributes. It did not test incoming components for identity. It relied on supplier certificates of analysis without ever establishing that the supplier's data was reliable, which 21 CFR 211.84(d)(2) requires before a firm can accept a certificate in place of its own testing.

The AI finding comes after all of that. The firm told investigators it had used AI agents to create drug product specifications, procedures, and master production and control records "to be in compliance with FDA requirements," and then used those documents without human review. Asked why it had not performed process validation, the firm said it was "not aware of the legal requirement, as the AI agent you used ... never told you it was required."

Who was running quality

PurOlea had four employees. None worked in regulatory affairs, quality or GMP manufacturing. The founder's background was five years as a sales representative and a marketing degree, with no prior experience in regulated manufacturing.

The firm used AI to produce the documents a quality unit would normally write and approve, and nobody at the firm was qualified to review them. That is the failure FDA cited.

Why the model didn't flag the gaps

A general-purpose model answers the question it is given, using the context it is given. It cannot see the loading dock, and nothing prompts it to raise a problem nobody asked about. Models are also trained on human preference ratings, and Anthropic's research on sycophancy found that raters and the preference models trained on them sometimes prefer a convincing incorrect answer over a correct one. The result is a model that tends to accept the user's framing.

Ask a chat window for an SOP and you get an SOP. It won't mention the insects in your production area. A tool built for regulatory work can do better if it has the relevant guidances, standards and product history in context and is designed to report gaps.

The tool still wasn't the deciding factor. Asked directly whether a facility with insects and filth in the production area meets CGMP, any current frontier model will say no. PurOlea never asked, and treated the generated SOP as a controlled record instead of a draft.

The standard FDA applied

The letter's instruction on AI is short: "If you use AI as an aid in document creation, you must review the AI generated documents to ensure they were accurate and actually compliant with CGMP."

That is the standard FDA already applies to a contractor's draft or a purchased template. FDA tied it to 21 CFR 211.22(c), which makes the quality unit responsible for approving or rejecting procedures and specifications that affect product quality. The letter does not restrict AI use. It holds the quality unit responsible for the documents, whoever or whatever drafted them.

Reading it as a case against AI

BioSpace's headline said the letter "signals new scrutiny for manufacturers," and some quality teams read it as a reason to pause AI tools. The letter doesn't support that reading. A firm with a functioning quality unit, document control, and staff who know Part 211 is in a different position from PurOlea, and the letter says nothing that should stop that firm from drafting with AI and reviewing the output.

For the mechanics of a warning letter and what a response has to contain, see the glossary entries on the FDA warning letter response, the Form 483 that usually precedes one, and AI in regulatory affairs. Published letters are searchable in our warning letter search, and the enforcement trend behind them is in What FDA Warning Letters Stopped Being About.

Sources

• FDA, Purolea Cosmetics Lab, 722591, 04/02/2026: fda.gov
• RAPS, FDA warns firm for inappropriate use of AI in drug manufacturing: raps.org
• BioSpace, FDA's first AI-focused cGMP warning letter signals new scrutiny for manufacturers: biospace.com
• MasterControl, FDA issues first warning letter for AI compliance: mastercontrol.com
• Anthropic, Towards Understanding Sycophancy in Language Models: anthropic.com