FDA’s Purolea Warning Letter: What AI-Assisted GMP Documentation Means for Manufacturers

1 day ago

5 min read

Artificial intelligence is rapidly moving into regulated manufacturing environments.

Manufacturers are increasingly using AI to assist with regulatory research, SOP drafting, specifications, quality documentation, and other activities that traditionally required significant human effort.

But where does the responsibility remain when AI is involved?

A recent FDA Warning Letter to Purolea Cosmetics Lab provides a useful enforcement example.

The Warning Letter, issued on April 2, 2026, addressed multiple CGMP deficiencies at the firm's drug manufacturing facility. One section specifically addressed the company's use of artificial intelligence to assist with pharmaceutical manufacturing documentation.

A July 2026 analysis published by Pharmaceutical Technology subsequently examined the case in the context of GMP accountability and AI-assisted documentation.

What did FDA find regarding AI?

According to the Warning Letter, Purolea told FDA investigators that it had used AI agents to create:

Drug product specifications
Procedures
Master production or control records

The company reportedly used AI to help comply with FDA requirements.

FDA's response is significant because it did not simply characterize the use of AI as prohibited.

Instead, FDA stated that when AI is used as an aid in document creation, the resulting documents must be reviewed to ensure that they are accurate and actually compliant with CGMP.

FDA cited the failure to perform that review as a violation of 21 CFR 211.22(c).

The FDA further stated that if the company resumes CGMP activities and uses AI to assist with activities such as developing procedures and specifications, AI-generated output or recommendations must be reviewed and cleared by an authorized human representative of the firm's Quality Unit.

AI does not replace the Quality Unit

This is perhaps the most important compliance lesson from the case.

Under the CGMP framework, the Quality Unit has defined responsibilities relating to the approval or rejection of procedures and specifications affecting drug product identity, strength, quality, and purity.

The introduction of AI into the drafting process does not change that responsibility.

Whether a document is written by:

a regulatory professional,
a manufacturing specialist,
an outside consultant,
an AI system, or
a combination of these,

the final GMP document still needs to go through the appropriate quality-system controls.

The technology used to generate the first draft does not determine who is accountable for the final document.

“The AI didn't tell us” is not a regulatory defense

Another important point in the Warning Letter involved process validation.

FDA investigators found that Purolea had not conducted process validation before distributing its drug products, as required under 21 CFR 211.100.

The company reportedly stated that it was not aware of the requirement because the AI agent it used had not told the company that process validation was required.

FDA nevertheless identified the process-validation deficiency.

This illustrates a fundamental principle:

AI output is not a substitute for regulatory knowledge or a manufacturer's responsibility to meet applicable requirements.

An AI system may overlook a requirement, misunderstand the applicability of a regulation, rely on an inappropriate source, or generate a generic answer that does not reflect the actual manufacturing process.

The manufacturer remains responsible for identifying and addressing the requirement.

Why AI-generated GMP documents require additional discipline

The challenge with AI-generated documentation is not limited to obvious factual errors.

A document can be grammatically correct, professionally formatted, and highly detailed while still being unsuitable for actual GMP use.

For example:

A specification may be based on the wrong source.

AI may identify a specification from an inappropriate or outdated reference without recognizing that the source is not applicable to the company's product.

An SOP may describe a generic process.

A procedure generated from general industry information may not reflect the company's actual equipment, validated parameters, sampling procedures, cleaning controls, or manufacturing environment.

A master production record may omit a control.

Because AI generates text based on patterns and available information, it may produce a document that appears complete while missing a site-specific requirement.

A regulatory assessment may miss applicability.

A general regulatory answer may not account for the specific formulation, claims, dosage form, manufacturing process, or intended use involved.

These are precisely the types of issues that qualified human review is designed to identify.

A practical AI governance model for GMP documentation

Manufacturers do not necessarily need to prohibit AI from all quality or regulatory activities.

Instead, companies should establish controls around how AI-generated content enters the quality system.

A practical model can include five stages:

1. AI-assisted drafting

AI may be used to organize information, generate an initial structure, or assist with preliminary drafting.

2. Qualified technical review

A qualified subject-matter expert evaluates the content against the company's actual process and applicable requirements.

3. Source verification

The reviewer verifies important regulatory and technical statements against authoritative sources and approved internal documents.

4. Quality Unit review and approval

Where the document falls within Quality Unit responsibilities, it should follow the established approval process regardless of how the initial draft was produced.

5. Controlled document management

The final version should be controlled through the company's established system, including appropriate version history, approval records, and effective-date controls.

This framework preserves the efficiency of AI while maintaining accountability within the existing quality system.

What about Part 11 and data integrity?

The use of AI also raises questions about electronic records and data integrity.

However, companies should avoid assuming that every AI interaction automatically becomes a Part 11 record.

The relevant question is how the electronic information is being used within the regulated process.

If an AI platform is simply used for informal brainstorming and the final GMP document is independently created, reviewed, approved, and controlled in the company's document-management system, the analysis may differ from a situation where the AI platform itself becomes part of the regulated record or approval workflow.

Companies should therefore evaluate:

What electronic records are relied upon?
Where are source documents stored?
What review evidence is retained?
Who approved the final document?
Can the company reconstruct the relevant review and approval process if requested during an inspection?

These questions should be addressed prospectively rather than after an inspection identifies a gap.

What should manufacturers do now?

For manufacturers, CDMOs, and other regulated organizations already using AI, the first step is visibility.

Identify where AI is currently being used in the quality and regulatory workflow.

Then classify those uses according to their potential impact.

An AI-generated internal brainstorming document does not carry the same risk as an AI-generated:

Product specification
SOP
Master production record
Validation protocol
Batch record instruction
CAPA-related document
Regulatory assessment affecting a GMP decision

The higher the potential impact on product quality, patient safety, regulatory compliance, or batch disposition, the more robust the review and approval process should be.

The key lesson

The Purolea Warning Letter should not be interpreted as an FDA ban on AI in GMP operations.

Instead, it provides a practical enforcement example of a much older principle:

Technology does not replace accountability.

AI can assist with drafting.

AI can help organize information.

AI can identify potential regulatory issues for further investigation.

But the manufacturer remains responsible for ensuring that the final document is accurate, appropriate, and compliant with applicable CGMP requirements.

For organizations adopting AI, the goal should not be to choose between technology and compliance.

The goal is to build a quality system in which AI-assisted work remains subject to qualified human review, source verification, Quality Unit oversight, and appropriate document control.