FDA’s first AI-focused cGMP warning letter signals new scrutiny for manufacturers

News Policy FDA’s first AI-focused cGMP warning letter signals new scrutiny for manufacturers July 16, 2026 | 3 min read | John Potter Twitter LinkedIn Facebook Email Print The FDAs enforcement action against Purolea Cosmetics Lab offers an early look at how the agency expects AI-generated work to fit within existing quality systems, with experts saying human oversight and quality controls remain essential. In April, the FDA issued what appears to be its first warning letter explicitly addressing the use of artificial intelligence in pharmaceutical manufacturing, signaling that AI-generated work remains subject to existing current good manufacturing practice requirements.The warning letter cited Purolea Cosmetics Lab for AI-related failures involving manufacturing documentation and quality unit oversight. The company has since ceased drug production.The inspection also identified additional cGMP deficiencies including failures in process validation, reinforcing that manufacturers remain responsible for complying with existing requirements regardless of how AI is used. Regulatory attorneys say the Purolea warning letter is best viewed as a traditional cGMP enforcement action rather than a new AI regulatory framework.“I consider the Purolea action to be a traditional action where AI is heavily involved,” James Boiani, an FDA regulatory attorney and member of the firm at Epstein Becker Green, told BioSpace in an email. “The Agency is expressing concern about lack of required oversight by a Quality Unit and all the violations that flow from that, which has been seen many times before.”What distinguishes the case, he said, is the underlying cause. “It wasn’t just people problems, it was problems caused by replacement of people with software, taking human oversight and critical thinking out of the loop.” Manufacturing FDA proposes streamlined requirements for ‘hub-and-spoke’ manufacturing models Through the proposal, the FDA could clear barriers to distributed manufacturing approaches intended to enhance emergency preparedness and supply chain resilience. July 14, 2026 · 2 min read · Nick Paul Taylor Read more The first enforcement signal for AI in GMP In its warning letter, the FDA devoted a section to the company’s use of AI in pharmaceutical manufacturing, distinguishing it from the other cGMP deficiencies the inspection had turned up. Investigators said Purolea used AI to generate standard operating procedures, drug product specifications and master batch production records without adequate review and approval by the company’s quality unit.The investigators also found that the company had not performed required process validation under FDA rules requiring written procedures and standards for all manufacturing efforts. When asked why validation had not been performed, company personnel responded that the AI agent had never identified the requirement.The FDA rejected that explanation, stating that reliance on AI does not relieve manufacturers of their cGMP responsibilities. AI-generated documents must still be reviewed for accuracy and compliance by qualified personnel in the quality unit. AI must fit existing cGMP expectations Artificial intelligence is already being used throughout pharmaceutical manufacturing. Applications range from drafting standard operating procedures and supporting deviation investigations to analyzing manufacturing data and monitoring equipment.“The fact that companies are using AI for GMP is not new,” said Vernessa Pollard, chair of FDA regulatory practice at law firm DLA Piper. “It just reinforces that when companies employ AI for these purposes, FDA expects them to validate the software and the algorithms for specific GMP uses.” She added that companies cannot rely exclusively on AI output to affirm compliance or safety and that “there needs to be a human in the loop.”Spencer Todd, CEO and co-founder of regulatory software company Dovetail, said manufacturers are adopting AI in different ways. “What you’re seeing is kind of a mix right now,” he said. “There’s purpose-built AI for technical documentation and traceability, and then there are broad-based consumer products that companies are using for efficiency and data management.”Todd added that the technology should be viewed as an efficiency tool rather than a replacement for technical expertise. “If it’s being generated by AI, it has to have an expert review it, because we’re just not there yet,” he said. Manufacturing Manufacturing issues prompt FDA to reject Hengrui-Elevar filings for third time An inspection of a facility making Elevar’s rivoceranib uncovered deficiencies, leading the FDA to reject a combination therapy that includes the drug. The rebuff is the third strike for the partners after the agency issued manufacturing-related denials for the other drug in the combo. July 15, 2026 · 1 min read · Nick Paul Taylor Read more What manufacturers should do next Pollard said companies are already asking practical questions about AI governance, including how to evaluate third-party AI tools, determine who should be authorized to use them and establish policies governing acceptable uses.Clients are also seeking guidance on validating AI-generated work before it becomes part of regulated manufacturing processes.The FDA issued broader guidance on AI in January 2025, but Pollard said she expects more detailed direction on GMP applications as AI adoption expands. “Given the proliferation of AI usage for GMP, I would expect to see cross-center AI for GMP guidance in the next year to two years, or product-specific AI guidance.” Twitter LinkedIn Facebook Email Print Artificial intelligence Manufacturing John Potter John Potter is a Michigan-based freelance journalist covering biotechnology, artificial intelligence and emerging technologies. He holds an MBA and writes about pharmaceutical innovation, AI and the business of emerging technologies.