On April 2, 2026, the US Food and Drug Administration issued Warning Letter 320-26-58 to Purolea Cosmetics Lab in Livonia, Michigan. It is a short document. But one section of it has sent a signal across the entire pharmaceutical industry that will not quickly be forgotten.
For the first time in the history of US drug regulation, an FDA warning letter contained a section explicitly titled: "Inappropriate Use of Artificial Intelligence in Pharmaceutical Manufacturing."
The company subsequently ceased drug production.
"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. Your failure to do so is a violation of 21 CFR 211.22(c)."
What Purolea did and why it was cited
Purolea Cosmetics Lab is a contract manufacturer producing homeopathic OTC drug products in Livonia, Michigan. Despite the word "Cosmetics" in the company name, it was operating as a drug manufacturer under 21 CFR Parts 210 and 211 and was therefore subject to cGMP requirements.
During an FDA inspection in late October 2025, investigators found that the company had used AI agents to create drug product specifications, procedures and master production or control records. These AI-generated documents were implemented without further human review by the quality unit.
The company's response to FDA's process validation finding made the situation significantly worse. When investigators noted that Purolea had not conducted process validation prior to distribution of its drug products, as required under 21 CFR 211.100, the company's reply was that it was not aware of the legal requirement because the AI agent it used had never told them it was required.
FDA's position on this was unambiguous. The agency cited two separate cGMP violations:
- 21 CFR 211.22(c) — failure of the quality unit to review and approve AI-generated documents before use
- 21 CFR 211.100 — failure to conduct process validation prior to distributing drug products
FDA did not write a new regulation to cite Purolea. It applied existing law. 21 CFR 211.22(c) has required quality unit review and approval of all procedures and specifications since the cGMP regulations were first published. AI-generated documents are not exempt from this requirement.
What 21 CFR 211.22(c) actually requires
21 CFR 211.22(c) states that the quality control unit shall have the responsibility and authority to approve or reject all procedures or specifications impacting on the identity, strength, quality, and purity of the drug product.
This regulation does not specify how a procedure or specification is created. It does not distinguish between documents written by a human, generated by software or produced by an AI agent. What it requires is that a qualified human reviews and approves the document before it enters the quality system.
FDA's position in this warning letter is therefore not new law. It is the application of existing law to a new technology. The quality unit cannot outsource its judgment to an AI tool. A qualified person must review AI-generated content, take responsibility for its accuracy and compliance, and document that review.
The two violations in plain language
Violation 1 — Unreviewed AI-generated documents
Purolea used AI to create quality documents and implemented them without quality unit review. This is not a question of whether the AI produced good or bad documents. It is a question of whether a qualified human reviewed them. The answer was no. That is the violation.
Violation 2 — Relying on AI to know the regulations
The process validation failure compounds the first violation in an important way. Purolea told FDA it did not know process validation was required because the AI agent never indicated it was. This is a fundamentally unacceptable compliance posture. AI tools do not replace the obligation to understand and comply with cGMP regulations. The responsible parties at a drug manufacturing facility are the people at that facility, not the software they use.
"Overreliance on artificial intelligence for your drug manufacturing operations was also documented during the inspection."
What this means for your quality system
This warning letter will not be the last FDA enforcement action to address AI in pharmaceutical manufacturing. The FDA and EMA jointly released Good AI Practice principles in January 2026, codifying expectations around traceability, validation, human oversight and lifecycle management for AI systems used in drug development and manufacturing. The regulatory direction is clear and it is accelerating.
Every pharmaceutical, biotech, medical device and contract manufacturing organisation using AI anywhere near a GxP workflow is now on notice. The Purolea enforcement action applies to a small homeopathic drug manufacturer, but the regulatory principle it establishes applies universally.
Here is what your quality system needs to address now:
- Inventory your AI use. Identify every AI tool being used in any GxP-adjacent activity — document creation, specification writing, SOP generation, batch record review, deviation investigation, regulatory submissions. Do this now, before an inspector asks.
- Establish review and approval workflows. Every AI-generated document that enters the quality system must be reviewed by a qualified human and approved through normal change control. The review must be documented. "The AI generated it" is not an approval.
- Do not use AI as a regulatory reference. AI tools can be wrong. They can be out of date. They can omit requirements. The responsibility to know and comply with cGMP regulations rests with the manufacturer, not the software. If your team is using AI to understand regulatory requirements, those outputs must be verified against the primary regulatory source.
- Include AI governance in your quality management system. AI use in GxP processes should be governed by a procedure that defines permitted uses, review requirements, documentation standards and change control obligations.
- Train your quality team. The people responsible for reviewing AI-generated content need to understand what they are reviewing. Rubber-stamping AI output is not compliance. Genuine critical review is.
FDA is not against AI in pharmaceutical manufacturing. The agency has said this explicitly. What it is against is using AI as a substitute for human judgement and quality oversight. AI can assist. A qualified human must still review, approve and take responsibility for every document that enters the quality system. That obligation is not new. It has always existed under 21 CFR 211.22(c). What is new is that FDA will now cite companies that fail to meet it specifically because of AI misuse.
The January 2026 FDA-EMA Good AI Practice principles
The Purolea warning letter did not arrive in a vacuum. In January 2026, FDA and EMA jointly published Good AI Practice principles covering expectations for AI systems used in drug development and manufacturing. Those principles cover traceability, validation, human oversight and lifecycle management. The enforcement action in April 2026 followed less than three months later.
The message from regulators is consistent and deliberate. They have published the expectations. They are now enforcing them. Sites that have not yet addressed AI governance in their quality management systems are operating with a compliance gap that FDA inspectors are now specifically looking for.
Key regulatory references
- FDA Warning Letter 722591 / 320-26-58, Purolea Cosmetics Lab, April 2, 2026
- 21 CFR 211.22(c) — Responsibilities of quality control unit
- 21 CFR 211.100 — Written procedures, deviations, and process validation
- 21 CFR Parts 210 and 211 — cGMP for finished pharmaceuticals
- FDA-EMA Good AI Practice principles, January 2026
Our AI in GMP Manufacturing course covers this enforcement action in full
Including the April 2026 Purolea warning letter, what 21 CFR 211.22(c) requires, Good AI Practice principles and what your quality system must do now. Data current to April 2026. Certificate on completion.