As public debate surrounding artificial intelligence continues to intensify, the press has announced that America’s “Great AI Freakout” has begun.[1] Beginning with warnings from industry insiders regarding the potential risks of AI systems, followed by calls for greater oversight and caution from AI executives, policymakers and commentators, AI skepticism is having a moment. More fundamentally, concerns about how AI development and deployment could be managed and controlled and its impact and risks mitigated have come to the forefront.
We have been addressing more prosaic concerns over the last few months, namely comments from food and agricultural processing clients that they intend to use AI, or have used AI, to analyze food safety problems and to develop the details of food safety plans or Hazard Analysis and Critical Control Point (HACCP) plans. Particularly for companies in the middle market and smaller segments, the use of AI may be viewed as providing an opportunity to avoid substantial investments of time and resources or the retention of outside consultants. While some may be concerned about the threat AI poses to human civilization, the deployment of AI which we have seen thus far would not be fully capable of understanding all the nuances of a manufacturing process, nor the very specific circumstances of a particular manufacturing facility or production line.
Thus far, it seems unlikely that AI tools or agents have the necessary knowledge or visibility to actual conditions in manufacturing processes that would permit the assessment of unique risks and to formulate strategies to mitigate those risks. In other words, competent human involvement remains essential. That AI is currently being used and deployed everywhere seems obvious, and its continued use and development seems inescapable (perhaps, too much so, in the opinion of some). Proponents of AI often suggest that one of its great benefits to humanity will be in the area of human health, such as pharmaceutical development. Presumably, AI will be helpful in food safety efforts.
Recently, the FDA issued what may be its first AI-focused warning letter to Purolea Cosmetics Lab.[2] The warning letter indicated that the company advised FDA investigators that the company had utilized AI agents to help the company comply with FDA regulations, including the creation of drug product specifications, procedures, and master production or control records. While the warning letter cited other violations regarding insanitary conditions and failure to test components or inputs, the letter specifically cited inappropriate use of Artificial Intelligence without adequately ensuring that authorized human representatives of the firm’s quality unit had reviewed and cleared any output or recommendations from an AI agent. Notably, the warning letter indicated that FDA investigators found that the company had not conducted process validation prior to distribution of drug products. Apparently, the company responded that it was not aware of this specific legal requirement because the AI agent utilized had “never told [the company] it was required.” The warning letter noted that the company had committed to ceasing drug production.
In this warning letter, the FDA has outlined what should be a rather obvious response to the deployment of AI: that there is currently no substitute for human audit, review and validation of the output of an AI agent. The Purolea warning letter clearly imposes responsibility on the “user” of the AI agent. In other words, it does not prohibit the use of AI, but rather it postures its use as a tool which must be used responsibly with intervention by properly trained humans. In this regard, the requirements for verification and validation – by humans – are not new and are regularly applied and enforced.[3] Companies therefore cannot rely on AI-generated recommendations without appropriate human review or assume that an AI tool’s failure to identify a regulatory requirement relieves them of compliance responsibilities. Unwarranted reliance on AI is also unlikely to be an effective defense to the application of the “Park Doctrine” or “Responsible Corporate Officer Doctrine.”[4]
The warning letter reinforces that the FDA does not prohibit the use of AI in Current Good Manufacturing Practice (CGMP) or other compliance-related activities. Rather, the FDA expects companies to maintain human oversight of AI-assisted work and to ensure that procedures, specifications, recommendations and other AI-generated outputs are reviewed and approved by qualified personnel before being relied upon for regulatory, quality assurance or food safety purposes.[5]
2026 has been a year of challenges in the food safety landscape in the United States. Good tools that support better food safety are to be welcomed. However, in the midst of the “Great AI Freakout,” food and agricultural processing companies should adopt a standard operating procedure or similar governance framework describing how AI may be used responsibly within the organization in support of food safety and compliance activities.
For assistance evaluating the use of AI in food safety and compliance functions or developing an appropriate policy or standard operating procedure, please contact Tim Dietrich or any attorney in Barley Snyder’s Food & Agribusiness Industry Group.
—
[1] Amrith Ramkumar, et. al., “America’s Great AI Freakout Has Begun,” Wall Street Journal, September 11, 2026. ; Andrew Ross Sorkin, “The A.I. Freakout,” New York Times DealBook, September 14, 2026.
[2] FDA Warning Letter to Purolea Cosmetics Lab, April 2, 2026.
[3] See, for example, 21 CFR § 117.155, 117.160, 117.165, 117.170 and, specifically, 117.180, which requires that a preventative controls qualified individual or a qualified auditor must “oversee” certain compliance activity. For an example of guidelines for validation which may be relevant to the recent outbreak of Cyclospora in leafy greens, see “Guidelines to Validate Control of Cross-Contamination during Washing of Fresh-Cut Leafy Vegetables.”
[4] See United States v. Park, 421 U.S. 658 (1975).
[5] See the Purolea warning letter, page 4.
