180 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 30 August 2026
← The incident library
HD-INC-158
Pharmaceutical manufacturing · United States · 2026 · Overreliance on unreviewed AI-generated regulated documentation

A US drug manufacturer told inspectors it had used AI agents to write the documents that define how its products are made, and said it had not known about a validation requirement because the AI had not raised it

By Ellie Harris · Filed FDA inspection 28 to 30 October 2025

Alleged: Purolea Cosmetics Lab developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

A US drug manufacturer told inspectors it had used AI agents to write the documents that define how its products are made, and said it had not known about a validation requirement because the AI had not raised it

What happened

It was reported that the US Food and Drug Administration inspected the drug manufacturing facility of Purolea Cosmetics Lab in Livonia, Michigan from 28 to 30 October 2025, and issued a warning letter to the firm on 2 April 2026. The letter runs through significant violations of current good manufacturing practice for finished pharmaceuticals: insanitary conditions under section 501(a)(2)(A) of the Federal Food, Drug, and Cosmetic Act, a failure to test finished products for microbiological attributes under 21 CFR 211.165(b), a failure to test components under 21 CFR 211.84(d), quality unit failures under 21 CFR 211.22, and the marketing of unapproved new drugs under sections 301(d) and 505(a). The products were sold as Dermveda Extra Strength Shingles Relief and Dermveda Extra Strength Ultra Genital Herpes Relief. One section of the letter carries a heading on the inappropriate use of artificial intelligence in pharmaceutical manufacturing.

It was reported that the letter records what the firm told investigators during the inspection. It had been using AI agents to help it comply with FDA regulations, and it had used AI to create drug product specifications, procedures and master production or control records. The FDA found the firm had not reviewed those AI-generated documents to make sure they were accurate and actually complied with current good manufacturing practice, and that this violated 21 CFR 211.22(c). Investigators also found the firm had distributed drug products without carrying out the process validation required by 21 CFR 211.100, and told it so during the inspection. According to the letter the firm answered that it had not been aware of the legal requirement, because the AI agent it used never told it that validation was required. The letter states that if the firm resumes drug production and uses AI in activities covered by good manufacturing practice, such as developing procedures and specifications, any output or recommendation from an AI agent is to be reviewed and cleared by an authorised human representative of the firm’s quality unit, under section 501(a)(2)(B) of the Act. The firm was given 15 working days to respond. A compliance consultant told NutraIngredients in June 2026 that this was the fourth FDA warning letter to mention artificial intelligence and the first to set out in detail how the technology can go wrong inside a quality system.

What an auditable version would have shown

A specification tells the people making a drug what that product has to be and how it is meant to be made. If AI helped write it, an inspector should be able to see who checked the work before anyone relied on it. The same applies to procedures and production records. Keep the AI draft, record what information it was based on, and record the person in the quality unit who reviewed and approved the final version. That would also make it possible to check what this case turns on: whether the documents covered all of the regulatory requirements they were supposed to cover, rather than only the ones the AI happened to identify.

Where the gap was

It was reported that the FDA found the firm had used AI to produce specifications, procedures and production records without properly checking them for accuracy or compliance. That is the work a ConstraintGate does: the AI can draft the document, but it stays a draft until someone qualified has reviewed and approved it. A ConductRecord keeps the AI output and the human approval together, so there is evidence of who checked what. There was another problem here as well. The firm told inspectors it did not know process validation was required because its AI agent had never mentioned it. AI can miss things. It can leave out a regulation, misunderstand one, or simply never raise it. The firm still needs its own way of checking that every requirement has been covered.

What governance should have looked like

It was reported that the FDA told the firm that future AI output used in manufacturing activities should be reviewed and cleared by an authorised person in its quality unit. That is a fairly simple control. AI can help write the document, but a qualified person checks it against the actual regulations before it is used. The review is recorded, the person approving it is named, and anything the AI missed can be picked up before the document reaches production. The wider FDA warning letter covered problems beyond AI, but this part of the case shows why AI cannot become the firm’s source of truth for compliance. If the model fails to mention a requirement, the requirement still exists.

Failure Pattern: the documents that define how a regulated product is made were created with AI and issued without a qualified person checking them for accuracy or compliance, and the absence of a legal requirement from the output was taken for its absence in law.

Governance Principle: where an AI system drafts a document that a regulated process depends on, a qualified person should review and clear it before it takes effect, and that clearance should be recorded against the document.

The reference implementation of ConstraintGate and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Purolea Cosmetics Lab could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →