FDA’s first AI warning letter

The Food and Drug Administration has tied a drug-manufacturing warning letter to artificial intelligence for the first time, telling one company that AI-written records still need human review. (raps.org) The letter went to Purolea Cosmetics Lab in Livonia, Michigan, on April 2, 2026, after an October 2025 inspection of its facility. FDA said the company used AI agents to draft specifications, procedures, standard operating procedures, and master production and control records. (gmp-compliance.org) FDA tied that conduct to 21 CFR 211.22(c), a quality-unit rule that requires written procedures to be followed and production records to be reviewed for errors. The agency said AI can be used as a drafting tool, but the finished documents still have to be checked for accuracy and compliance. (accessdata.fda.gov) (redica.com) Current good manufacturing practice is the rulebook for how drug plants make, test, document, and release products. In this case, FDA treated weak AI controls as part of a broader manufacturing failure, not as a separate software policy dispute. (fda.gov) (outsourcedpharma.com) The inspection record described a company that leaned on AI in place of basic quality knowledge. One example cited by industry write-ups: firm personnel said they did not know process validation was required because the AI system had not surfaced that requirement. (propharmagroup.com) (leucine.ai) FDA’s letter also covered more familiar drug-quality problems. Reports summarizing the letter said investigators found failures involving process validation under 21 CFR 211.100, batch review, microbiological testing, and facility conditions, alongside marketing claims for homeopathic products aimed at shingles and genital herpes. (raps.org) (changeflow.com) That makes the AI section less a ban on automation than a warning about traceability. If a company cannot show who reviewed an AI draft, what was changed, and why the final record meets drug-manufacturing rules, FDA can treat that gap as a current good manufacturing practice violation. (clarkstonconsulting.com) (complianceg.com) Regulators have been signaling that approach for years by applying existing quality rules to new tools rather than waiting for AI-specific regulations. This letter puts that position into enforcement language: machines can help write the paperwork, but a manufacturer’s quality unit still owns the paperwork. (fda.gov) (gmp-compliance.org)