Can you use AI to write SOPs and other GMP documents?

1What FDA’s warning letter found.

FDA inspected Purolea Cosmetics Lab in Livonia, Michigan, from 28 to 30 October 2025 and issued Warning Letter 320-26-58 on 2 April 2026. The firm told investigators it had used AI agents “to create drug product specifications, procedures, and master production or control records to be in compliance with FDA requirements.” FDA’s response: “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).”¹

The letter also documents “overreliance on artificial intelligence”. The firm had distributed product without process validation, which 21 CFR 211.100 requires, and when told so replied that it was not aware of the requirement because the AI agent had never told it. If the firm resumes production and uses AI for work such as procedures and specifications, “any output or recommendations from an AI agent must be reviewed and cleared by an authorized human representative of your firm’s QU”.¹ The letter does not cite the use of AI itself. It cites the missing review.

2Who must approve a GMP document.

The rules for approving a document don’t change because AI drafted it.

Chapter 4 also expects SOPs to be written in an “imperative mandatory style”, and documents to be regularly reviewed and kept up to date.³ FDA accepts electronic signatures with appropriate controls in place of the handwritten signatures 211.186(a) specifies for master records.⁶

3What the EU drafts say about AI-made documents.

The revised Chapter 4 that the European Commission and PIC/S published for consultation in July 2025 addresses AI directly. Draft paragraph 4.24 reads: “The accountability for the integrity of documents, records or (raw) data produced or processed with artificial intelligence or any other automatic means (e.g. validation scripts) rests with the regulated user.” Paragraph 4.25 asks for support by any automatic means, artificial intelligence included, to be part of the pharmaceutical quality system “regardless of the service located on premise or as a hosted service”.⁷ The companion draft Annex 22 says generative AI and large language models should not be used in critical GMP applications, and that in non-critical ones qualified, trained people are responsible for the output.⁸ Both are drafts. On 6 November 2026 EudraLex Volume 4 still lists the January 2011 Chapter 4 and no Annex 22.⁹

EMA’s final reflection paper on AI says something similar about medicinal product information. AI used to draft, edit, translate or review those documents “should be used under close human supervision”, because generative language models “are prone to include plausible but erroneous or incomplete output”.¹⁰

4What to check in an AI draft.

The approval route is the same as for any SOP, but an AI draft fails in its own ways. Check:

• Requirements against the regulation itself. Purolea told FDA it had not known process validation was required because its AI agent never said so.¹
• The procedure against what the site actually does. Production procedures must be designed to assure identity, strength, quality and purity, and must be followed as written.⁴
• The content against the marketing authorisation and product specification file.³
• Every citation and reference. NIST’s Generative AI Profile warns that generated output can include confabulated citations that make wrong content look justified.¹¹
• Training for the people who will work to it. Each person must be trained in the written procedures that relate to their functions.¹²

5How it runs in Seal.

Neil can turn a procedure into a process: give it an SOP and it drafts a process configuration in Seal, with the steps, checks and approval gates mapped to the source procedure. Your team checks the intended use, resolves ambiguities, tests the configuration and releases it under change control.

When an SOP changes, Neil can work through the change. It reads the previous and proposed revisions, the current process configuration and the linked training and change-control records, then drafts a source-mapped change summary and proposed process updates. The process and quality owners assess the impact, training and verification needed. In Word, Neil changes the file only once you accept its edits.