When AI Writes the GMP Document: What the FDA's Purolea Warning Letter Means for Beauty Brands

Sep 22

5 min read

Desk with laptop showing AI-to-document graphic, GMP paper, and FDA regulations books beside headline on FDA's Purolea warning letter.

Artificial intelligence is quickly becoming part of everyday business operations — from product research and regulatory searches to SOP drafting and quality documentation.

A recent FDA Warning Letter involving a Michigan drug manufacturer highlights an important boundary:

AI can help create a document. It cannot take responsibility for the document.

On April 2, 2026, FDA issued Warning Letter 320-26-58 to Purolea Cosmetics Lab, a Livonia, MI-based manufacturer, following a three-day inspection in late October 2025. Among a long list of CGMP deficiencies, FDA devoted a specific section to the company's use of artificial intelligence — reportedly the first time FDA has explicitly addressed AI-generated documentation inside a CGMP warning letter. The case has since drawn analysis from outlets including Pharmaceutical Technology, RAPS, and BioSpace.

One clarification worth making up front

Despite the name, Purolea wasn't a cosmetics brand that got pulled into drug territory by a specific claim. The company was already manufacturing homeopathic OTC drug products — some labeled to treat shingles and genital herpes — which is why 21 CFR Parts 210 and 211 applied to begin with. The AI issue was one part of a much longer list of findings: unsanitary manufacturing conditions, no microbiological testing of finished product, no identity or purity testing of incoming components, unapproved drug claims, and no process validation. The company has since ceased drug production at the facility.

So this isn't a case study in "how a cosmetic accidentally became a drug." It's a case study in what happens when AI-generated documents enter a quality system with no one checking them — which is exactly the risk beauty brands need to think about before they cross that line, not after.

What happened with the AI documentation

According to the warning letter, Purolea told investigators it had used AI agents to help create:

Drug product specifications
Procedures
Master production and control records

FDA did not say that using AI for these tasks was inherently prohibited. What it said was that any AI output used to support CGMP activities — like developing procedures and specifications — must be reviewed and cleared by an authorized member of the firm's Quality Unit, citing 21 CFR 211.22 and 211.100.

The company hadn't done that. The documents went into use without Quality Unit review.

AI didn't flag the requirement. That's not a defense.

One of the more striking details involves process validation. Investigators found the firm hadn't conducted process validation before distributing its drug products — a clear requirement under 21 CFR 211.100.

The company reportedly explained that it wasn't aware of the requirement because the AI agent it used had never told them it existed.

FDA didn't accept that as an excuse. The lesson is straightforward:

An AI system can miss a regulatory requirement. The manufacturer's responsibility doesn't disappear because the AI missed it.

This matters more, not less, as AI gets better at producing documents that look complete and professional. A polished SOP is not the same thing as a compliant SOP.

Why this matters to the beauty industry specifically

Purolea itself wasn't a cosmetic-to-drug crossover — but the boundary it sits on either side of is exactly the one many beauty brands operate near.

A product marketed as a cosmetic in Korea or elsewhere can be regulated very differently in the U.S. depending on ingredients, claims, and intended use. Sunscreens are the classic example. Acne-treatment products, anti-aging products with drug-like claims, and certain "medical device"-adjacent beauty tech can all fall within the U.S. drug or device framework.

That means a company developing what looks like a conventional beauty product may, without fully realizing it, be taking on U.S. drug CGMP obligations — including how it creates and controls specifications, procedures, and batch records.

For Korean beauty manufacturers and brands entering the U.S. market, this distinction becomes especially relevant when adapting formulas, claims, labels, and manufacturing documentation for U.S. regulatory purposes — often the exact kind of work companies are now turning to AI to help speed up.

The real risk isn't AI. It's uncontrolled reliance on AI.

AI can be genuinely useful for:

Creating an initial document structure
Organizing information
Comparing regulatory requirements across markets
Flagging potential gaps for further investigation
Supporting internal drafting

The risk starts when an AI-generated document is treated as if it's already verified.

An AI-drafted procedure might describe a manufacturing step that doesn't match the actual validated process. A specification might rely on an outdated or incorrect source. A master production record might quietly omit a required control. A regulatory gap analysis might miss a requirement simply because the AI model wasn't trained on it or didn't surface it.

The document can still look perfectly professional. That's the trap.

So the question isn't:

"Did AI write this?"

It's:

"What happened after AI wrote it?"

A practical control model

For AI-assisted GMP documentation, a simple workflow provides a much stronger compliance foundation:

1. AI Draft — Use AI to support initial drafting or organization.

2. Qualified Human Review — A qualified subject-matter expert reviews the content against applicable requirements and the company's actual processes.

3. Source Verification — Cross-check regulatory requirements against authoritative sources, approved specifications, validated procedures, and other controlled references.

4. Quality Unit Approval — Where applicable, the document follows the same Quality Unit review and approval process required for any other GMP-controlled document.

5. Document Control — The final approved version enters the company's established document-control system with proper version history and approval records.

This lets a company benefit from AI's speed without treating AI output as regulatory authority.

Questions beauty companies should be asking now

Where are we currently using AI in regulatory or quality work?
Which AI-generated documents could affect product quality or compliance status?
What source documents are being used to verify the output?
Who is qualified to review the content?
Does the document require Quality Unit approval?
How is the final approved version controlled?
What evidence of review and approval do we retain?

The answers will differ by product category and quality system maturity — a marketing draft and a GMP master production record shouldn't be treated the same way.

The bigger lesson for K-Beauty

As K-Beauty brands keep expanding into the U.S., speed matters — but so does knowing exactly when a product crosses into a more heavily regulated category.

AI can make it faster to draft a specification, SOP, regulatory assessment, or quality document. It doesn't make the underlying regulatory obligation disappear, and — as Purolea shows — it won't necessarily tell you a requirement exists in the first place.

The Purolea warning letter is less a statement about whether companies should use AI, and more a statement about how AI-assisted work has to sit inside an existing quality system. For beauty companies operating near the cosmetic/drug boundary, that distinction is only going to matter more.

AI can accelerate the work. Human expertise, source verification, and quality oversight still have to own the result.

Sources

>>>>Click Here and Start Your Own K-Beauty Products Now

Ready to launch in the U.S.? Contact us today.

CONTACT US

Office: 1-909-229-8249

Email: hello@theglowbridge.com