In April, the FDA sent the letter to Michigan-based Purolea Cosmetics Lab, who makes homeopathic drugs, which have been under scrutiny as a regulatory category by the agency for more than a decade. The FDA inspected the company and found Good Manufacturing Practices (GMP) and disease claim violations, including for the products Dermveda Extra Strength Shingles Relief and Dermveda Extra Strength Ultra Genital Herpes Relief.
The company also used “AI to create drug product specifications, procedures, and master production or control records to be in compliance with FDA requirements.”
The FDA wrote that “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… For example, the FDA investigators found that you had not conducted process validation prior to distribution of your drug products, as required under 21 CFR 211.100, and informed you as such. 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.”
This was the fourth FDA warning letter to mention artificial intelligence. According to Asa Waldstein, founder of Apex Compliance, it is the first warning letter to go into detail about best practices and the potential flaws when using this technology in quality systems. Waldstein published his commentary on last week’s Warning Letter Wednesday (WLW) website, where he explores the FDA letters and enforcement trends.
“AI is helpful for strategy,” said Waldstein, who is also chair of the American Herbal Products Association’s (AHPA) Technology & AI Innovation Committee. “For example, it should help the user understand what should be on a quality document and why. Pro tip: The more information and examples you can give the AI, the higher the chances are that the output will be what you are looking for.”
He added that AI is also good at repetitive tasks, such as making global changes to large volumes of documents, though human verification is needed.
“Expect the FDA or any auditing group to scrutinize important quality documents that have been made with AI unless they are human-rationalized and signed off,” Waldstein noted. “Essentially, staff should be able to defend the ‘what and why’ of the quality documents rather than pushing the responsibility to AI.”
‘Never 100% reliable’
It can be risky to use AI unless the tool and process are validated and documented. The foundation of GMPs is that all processes and procedures should be repeatable and verified, Waldstein wrote.
“We can apply the ‘trust but verify’ adage to all aspects of GMPs, including Master Manufacturing Records (MMRs), Standard Operating Procedures (SOPs), and product specifications,” Waldstein said. “This applies to both human- and AI-developed systems.”
He likened it to some Quality Management Systems when they initially went digital. These systems allowed companies to cut down on paperwork and increase efficiency, but they also required validation to ensure they were accurate, reliable and maintained data backups.
“AI can be a black box, meaning it’s not a standard algorithm and sometimes acts in surprising ways, including the possibility of hallucinations (making up content),” Waldstein said. “Large AI labs such as OpenAI and Anthropic continue to improve AI safety and reliability, but AI should never be relied on 100%, meaning human-in-the-loop oversight is essential to any AI quality system review.”
