A Michigan drug maker let AI agents write its manufacturing records, shipped product without validating the process, and told the FDA the AI never mentioned validation was required. That last line is now part of regulatory history.
On April 2, the FDA sent Purolea Cosmetics Lab of Livonia what compliance lawyers are calling the first warning letter to cite AI misuse as a named cGMP deficiency. The firm used AI agents to generate drug product specifications, procedures, and master production records, the documents that define how a drug gets made.
The FDA cited 21 CFR 211.22(c): the company relied on AI-generated procedures and specifications without review and approval by authorized quality unit personnel. It also cited 211.100, distributing drug products without process validation. The reason given, per the letter, was that personnel didn’t know validation was required because the AI agent didn’t identify it.
Here’s why this one matters beyond a single small manufacturer. The FDA’s position reads as a marker for the whole industry: responsibility for cGMP compliance can’t be delegated to a tool, and a human-in-the-loop is the expectation, not a nice-to-have. Every pharma quality team now piloting AI for SOPs and batch records just got a citable precedent for what happens when the human step gets skipped. The agency built this case on existing regulations, no new AI rule needed.
The letter doesn’t list a fine. It lists a precedent.