Only one of 3,011 FDA warning letters issued from January 2022 to September 2026 treated a firm's own use of AI as a violation, a new paper reports. That letter, sent to US firm Purolea Cosmetics Lab on 2 April 2026, says AI output must be reviewed and cleared by the quality unit.
The paper appeared on 31 October 2026 and is now being promoted through a press release. We read the FDA letter and the paper's data. The core finding holds, but "earliest" is narrower than it sounds, and the authors work for a company that sells a regulatory database. For Irish quality teams, the EU rule on AI in GMP is still a draft.
- The letter: FDA warning letter 320-26-58, dated 2 April 2026, from the Center for Drug Evaluation and Research (CDER) to Purolea Cosmetics Lab in Livonia, Michigan.
- The AI finding: FDA says failing to review AI-generated documents is a violation of 21 CFR 211.22(c), an existing rule. It sits alongside insanitary conditions, missing testing and unapproved drug findings.
- The study: 7 of 3,011 letters (0.23%) referred to AI: 3 device, 3 tobacco and 1 drug. Only the Purolea letter treated a firm's use of AI as a violation.
- Who ran it: George Kwiecinski and Reet Nandy, who both work for KeyPedia, a commercial regulatory database. They declare this in the paper.
- EU position: Annex 22 on AI is not in the EU GMP Guide yet. EMA's inspectors group has a Q4 2026 target to give the European Commission a final text.
What does the FDA letter actually say about AI?
FDA inspected the Livonia site from 28 to 30 October 2025. According to the letter, the firm told investigators it used AI agents to help it comply with FDA rules, including to create drug product specifications, procedures and master production or control records. The name of the tool is redacted, and the letter does not say what kind of model it was.
The passage appears under the heading "Inappropriate Use of Artificial Intelligence in Pharmaceutical Manufacturing". FDA writes: "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)."
That rule does not mention AI. It says the quality control unit "shall have the responsibility for approving or rejecting all procedures or specifications impacting on the identity, strength, quality, and purity of the drug product."
FDA also gives an example of what it calls overreliance. Investigators found the firm had not carried out process validation before distributing its products. FDA records the reply: "You replied that you were not aware of the legal requirement, as the AI agent you used (b)(4) , never told you it was required."
The letter does not ban AI. It says that if the firm resumes production and uses AI for CGMP activities, "any output or recommendations from an AI agent must be reviewed and cleared by an authorized human representative of your firm's QU".
Was AI the main problem at this site?
No. AI is one part of a wider letter. FDA says investigators observed "insects, filth, leaves, and clutter in several areas" and that homeopathic products were released without microbiological testing. It also cites inadequate component testing and quality unit failures. It says two labelled products, Dermveda Extra Strength Shingles Relief and Dermveda Extra Strength Ultra Genital Herpes Relief, are unapproved new drugs.
FDA acknowledges the firm's commitment to stop drug production and distribution, and notes that products within expiry remain in US distribution. A warning letter sets out FDA's findings and gives the firm 15 working days to respond. We found no public statement from the firm and no close-out letter on the FDA page when we checked on 6 November 2026.
Does the "earliest" claim stand up?
Partly, and only in the terms the authors use. The press release headline reads "Independent Study of FDA Correspondence Finds Earliest CDER Warning Letter Citing AI Use". The paper says it is "the earliest CDER warning letter we identified that addresses a firm's use of AI, and the only one in the period reviewed". So it is the earliest in a set of one.
There are limits. The search began on 1 January 2022, so older letters were not checked. The seven search terms did not include product names such as ChatGPT. Letters issued but not yet posted were missed. One author screened and coded the results with help from a large language model, with no second coder. The authors also say that with 7 letters "no trend can be established".
It is checkable, because the paper's supplementary file lists all 11 letters that matched a search term, with FDA's wording. We read the file and the Purolea letter, but we did not repeat the search of all 3,011 letters. The claim also does not mean Purolea was the first FDA warning letter to mention AI. Law firm Morgan Lewis wrote in April that it was "not the first FDA Warning Letter related to AI", since earlier letters dealt with the regulatory status of AI systems themselves.
| FDA programme and number of letters | How AI appears |
|---|---|
| Device (3): WAVi, Exer Labs, SeniorLife Technologies | Unvalidated software with AI capabilities, or AI-based screening and diagnosis claims |
| Tobacco (3): North Disposable, Green Rush, Hellosynixvape.com | "AI chat" listed as a vape feature that may appeal to youth, outside the cited violations |
| Drugs (1): Purolea Cosmetics Lab | Firm's use of AI to write GMP documents without review, cited under 21 CFR 211.22(c) |
Who ran the study, and do they sell anything?
The authors are George Kwiecinski, listed at the University of Scranton, and Reet Nandy, listed at New York University. The paper appeared in Discover Pharmaceutical Sciences, which Springer Nature describes as a peer-reviewed, open access journal. It was received on 1 July and accepted on 19 October 2026.
The competing interests statement reads: "Both authors work for KeyPedia (legal entity: Global Key Solutions Corp.), which develops a commercial database of FDA and other regulatory records." KeyPedia's website offers alerts on warning letters and demos of its platform. The authors report no specific grant or external funding.
The press release headline calls the study "independent". Readers should weigh that against the declared interest. The EIN Presswire site refused our request for the full release, so this report relies on the paper, its data and the FDA letter. To the authors' credit, their conclusion is restrained: "An enforcement precedent exists, but it rests on existing requirements, not new ones."
What are FDA and EMA saying about AI in GMP?
In the US, the Purolea letter applies existing CGMP rules. FDA's January 2025 guidance on AI to support regulatory decision-making for drugs and biologics is still listed as a draft. In January 2026, FDA and EMA issued 10 Guiding Principles of Good AI Practice in Drug Development. The document covers manufacturing as well as other phases, and says AI should adhere to standards "including Good Practices (GxP)". It sets principles, not binding GMP rules.
In the EU, the European Commission consulted on a new Annex 22, a revised Annex 11 and a revised Chapter 4 from 7 July to 7 October 2025. The Annex 22 draft covers AI models in critical applications with direct impact on patient safety, product quality or data integrity. It covers static, deterministic models only.
The draft says it "does not apply to Generative AI and Large Language Models (LLM), and such models should not be used in critical GMP applications." In non-critical uses, it says qualified and trained staff should "always be responsible for ensuring that the outputs from such models are suitable for the intended use".
That wording may change. An EMA report published on 2 October 2026 says consultation comments supported allowing LLMs and generative AI in GMP applications, and that the drafting group has discussed widening the scope. It says "concerns remain" about the control strategy needed. The report's conclusion says there "may be viable pathways" for such technologies, subject to appropriate controls. The workshop gathered mainly industry input, so these are not settled EMA positions.
| Document | Status |
|---|---|
| FDA warning letter to Purolea, 2 April 2026 | Addressed to one firm; applies existing 21 CFR Part 211 |
| FDA draft guidance on AI for regulatory decisions, January 2025 | Draft, not for implementation |
| FDA and EMA Guiding Principles, January 2026 | 10 principles for good practice |
| EU GMP Annex 22, draft | Consultation closed 7 October 2025; not listed in EudraLex Volume 4 |
| EU GMP Annex 11 and Chapter 4 | January 2011 versions still listed; revisions consulted on |
What does this mean for Irish GMP and quality teams?
The letter and the paper do not mention Ireland, and we found nothing linking Purolea or its products to the Irish market. The link is regulatory. The HPRA carries out GMP inspections of manufacturers here, under the EU GMP Guide. Irish sites that supply the US can also face FDA inspection.
Our guide to FDA warning letters in Ireland records two letters to manufacturing sites on the island from January 2021 to September 2026. The June 2026 letter to Genzyme in Waterford also cited quality unit oversight under 21 CFR 211.22, although it is not among the study's AI matches. A search of hpra.ie by Priya Life Science did not find HPRA guidance specific to AI in GMP.
Three practical points follow from the documents themselves:
- AI-drafted GMP documents still need quality unit approval. The authors say firms should also be able to show inspectors how that review is done.
- A tool's silence is not a defence. In the letter, FDA recorded the firm's explanation about process validation and still cited the requirement.
- Accountability stays with the firm. A key message from the EMA workshop was that "The regulated user remains accountable even when using third-party AI or cloud services."
Readers working with generative tools may also find our evaluation of the Irish Medtech GenAI roadmap useful. It covers medtech rather than medicines GMP.
Frequently asked questions
Was this the first FDA warning letter to mention AI?
No. The paper lists six letters before it, from the device and tobacco programmes, starting in 2023. It is the only one in the review that treated a firm's use of AI as a violation.
Has FDA banned AI in drug manufacturing?
No. The letter says AI output used in CGMP activities must be reviewed and cleared by an authorised member of the quality unit.
Is EU GMP Annex 22 in force?
No. It is not listed in EudraLex Volume 4. It remains a draft under revision.
Does the study show that FDA enforcement on AI is rising?
No. The authors say 7 letters cannot establish a trend, and that the results do not measure how widely industry uses AI.
I use a product named in the letter. What should I do?
This article is not medical advice. The letter concerns products in US distribution. Nobody should change treatment for shingles, genital herpes or any other condition without talking to their doctor or pharmacist.
Outlook
The date to watch is the end of 2026. EMA's Inspectors Working Group work plan sets a Q4 2026 target for giving the Commission a final Annex 22 text, in parallel with Annex 11 and Chapter 4. It also plans inspector training on the revised documents during 2026 to 2028. That is a target, and we found no published adoption date. The main question is whether generative AI stays excluded from critical GMP uses. Priya Life Science will update this report if the final text appears, or if the firm or FDA publishes a further response.
Sources
- US Food and Drug Administration, Warning Letter 320-26-58, Purolea Cosmetics Lab, MARCS-CMS 722591, 2 April 2026
- Kwiecinski G, Nandy R, Discover Pharmaceutical Sciences, Artificial intelligence in FDA regulatory correspondence, 2022 to 2026: a retrospective document analysis, with Supplementary File 1, 31 October 2026
- US Code of Federal Regulations, via Cornell Law School, 21 CFR 211.22, Responsibilities of quality control unit, read 6 November 2026
- US Food and Drug Administration, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products (draft guidance), January 2025
- US Food and Drug Administration and European Medicines Agency, Guiding Principles of Good AI Practice in Drug Development, January 2026
- European Commission, Stakeholders' Consultation on EudraLex Volume 4: Chapter 4, Annex 11 and New Annex 22, 7 July to 7 October 2025
- European Commission, Annex 22: Artificial Intelligence (consultation draft), July 2025
- European Commission, EudraLex Volume 4, Good Manufacturing Practice guidelines, read 6 November 2026
- European Medicines Agency, Report: Multistakeholder workshop on expert contributions to artificial intelligence guidance development (Annex 22), workshop 30 June 2026, report published 2 October 2026
- European Medicines Agency, The 3-year work plan for the Inspectors Working Group, January 2026 to December 2028, 2026
- Health Products Regulatory Authority, Guide to Information Held by the HPRA (MGT-G0016-12), 17 April 2026
- Morgan Lewis, FDA's Warning Letter Suggests Growing Scrutiny of AI Overreliance, 24 April 2026
- KeyPedia, company website, read 6 November 2026
Cover image: FDA Entrance, U.S. Food and Drug Administration, public domain, via Wikimedia Commons.
