Fifty years ago, the rock band Pink Floyd released the classic single, "Welcome to the Machine." Today, the title of this iconic song is an apt descriptor of our modern technological landscape. Artificial intelligence (AI), big data, and machine learning are transforming the commercial world, including the food industry. These tools have the potential to bolster risk management, compliance, and quality assurance efforts for food manufacturers.

According to the U.S. Food and Drug Administration's (FDA's) final rule, Current Good Manufacturing Practice, Hazard Analysis, and Risk Based Preventive Controls for Human Foods (i.e., the CGMP rule), which implements provisions of the Food Safety Modernization Act (FSMA), food manufacturers must create proactive safety plans to minimize food adulteration, among other requirements.¹ In addition to FSMA, food manufacturers are subject to extensive regulations by FDA with respect to food additives, an area that has been the subject of increased scrutiny under the new presidential administration.

Advances in AI offer promising opportunities to strengthen food manufacturers' internal compliance efforts with FDA requirements. While FDA has issued written guidances on AI, primarily for drug and medical device companies, its direct application to food safety is still evolving.

We will explore how AI might support food safety efforts and regulatory compliance.

Food Safety Plans Required by FSMA and FDA

In 2015, FDA promulgated the CGMP Rule to implement provisions of FSMA that impose internal compliance and quality assurance obligations on food manufacturers. Generally, the rule applies to domestic and foreign food facilities that are required to register under the Federal Food, Drug, and Cosmetic Act (FDC Act), which includes facilities involved in manufacturing, processing, packing, or holding food for human consumption in the U.S. (Note: The rule lists several entity-level exemptions, including farms, retail food establishments, restaurants, and certain nonprofit food establishments.)

FDA regulation requires food facilities to implement and follow a food safety plan that includes an extensive analysis of hazards and risk-based preventive controls to minimize or prevent the identified hazards. Specifically, a food safety plan must include the following elements.

Hazard analysis. A hazard analysis must consider known or reasonably foreseeable biological, chemical, and physical hazards in the facility's manufacturing or handling of human food. A hazard is any biological, chemical, or physical agent that could potentially cause illness or injury. Examples include pathogens; pesticide, drug, or heavy metal residues; allergens; and industrial chemicals. A hazard analysis involves assessing the severity of potential illnesses or injuries that would likely result if the hazard were present and the probability that the hazard would occur in the absence of preventive controls.

If a hazard analysis reveals that one or more hazards require a preventive control, then the food safety plan must identify the preventive controls that will be implemented, such as a risk-based supply chain program (described further below).

FDA's standard for determining whether a hazard requires a preventive control is whether a person knowledgeable about safe manufacturing, processing, packing, or holding of food would, in conducting a hazard assessment, conclude that a preventive control was needed.

Preventive controls. Preventive controls, when required, must be implemented to ensure that any hazard requiring a preventive control will significantly minimize or prevent food adulteration. Preventive controls include:

• Process controls: Procedures that ensure the control parameters of food being handled by a facility are met, such as cooking, refrigerating, and acidifying; pasteurizing; dehydrating/drying; or high-pressure processing of foods.
• Food allergen controls: Written procedures the facility must implement to control allergen cross-contact and ensure allergens are properly labeled on food packaging.
• Sanitation controls: Procedures, practices and processes to ensure that the facility is maintained in a sanitary condition to minimize or prevent hazards such as biological pathogens or food allergens.
• Other controls: Controls not described above but are also necessary to ensure that a hazard requiring a preventive control will be significantly minimized or prevented. 

Management and Oversight of Preventive Controls in Food Safety Plans

A company must continue to manage the preventive controls that it implements to minimize known hazards, including:

• Monitoring the implementation of the preventive controls.
• Verifying that preventive controls are executed in accordance with the food safety plan.
• Implementing corrections and corrective actions when preventive controls are not used according to the food safety plan.

Also, if a facility implements a risk-based supply chain program in response to a hazard analysis, then the facility must only accept raw materials and other ingredients necessitating a preventive control (i.e., materials that commonly carry associated hazards) from approved suppliers. The facility must take steps to confirm the consistent use of preventive controls and document this verification process.

Leveraging AI in Food Safety Plans

Ensuring compliance with FSMA-mandated food safety plans requires continuous monitoring, verification, and corrective actions—tasks that can be complex and resource-intensive. As food supply chains become increasingly globalized and intricate, the need grows for innovative solutions aimed at enhancing efficiency while maintaining regulatory compliance.

Advances in machine learning offer promising possibilities to improve not only food safety but also food safety plans required under FSMA. In particular, there may be opportunities for food facilities to incorporate machine learning applications into their procedures for managing and overseeing preventive controls. For example, as early as 2010, facilities were utilizing machine learning to develop sensor-based devices (e.g., spectroscopy) to detect fecal contamination in leafy greens during manufacturing processes. In addition, quality management studies have identified and ranked facility-level factors associated with post-pasteurization contamination of milk.² Similarly, machine learning applications have helped develop electronic, acoustic, and ultrasonic sensor technologies that are capable of monitoring the contamination and cleaning of food processing equipment. Food facilities could incorporate similar machine learning applications in their own efforts to monitor and verify that preventive controls are being executed effectively and consistently, as required by the CGMP rule.

Some manufacturers are incorporating machine learning applications into their own processes. For example, Tyson Foods has explored AI-powered computer vision systems to monitor hygiene compliance and detect foreign objects in production lines.³ Separately, IBM has created an AI-powered blockchain system for food manufacturers and retailers that purports to promote traceability in the food supply chain, thereby allowing them to track food products in real time.⁴ Such technologies could play a valuable role in developing a supply chain program as part of a facility's food safety plan.

Food Additives and AI

According to the FDC Act, a food additive is any substance of which the intended use results or may reasonably be expected to result, directly or indirectly, in its becoming a component or otherwise affecting the characteristics of any food.⁵ FDA must authorize a food additive before it is incorporated into food that is sold for public consumption or used in a manner that was not previously approved by the agency.⁶ An approval is difficult to obtain, as the pre-market review process is rigorous and requires the petitioner to provide extensive evidence to persuade FDA that the food additive is safe for consumption according to the proposed conditions of use.

A food manufacturer may use a food additive without FDA approval if it has concluded that the use of the additive can meet the standard for a food that is "generally recognized as safe" (GRAS). To satisfy the GRAS standard, a food additive must be generally recognized among qualified experts as having been adequately shown to be safe under the conditions of intended use.⁷ A manufacturer is encouraged (but not required) to notify FDA of its intent to incorporate a food additive into its food production and determinations that the additive satisfies the GRAS standard.⁸ However, FDA can subsequently determine that a food additive does not meet the GRAS safety standard, in which case the agency may seek to prevent continued use in food production by:

• Issuing a public warning letter to companies that manufacture or distribute the additive.
• Issuing a public alert about the risks of consuming foods containing the additive.
• Taking enforcement action to stop distribution of the food ingredient and foods containing it because they contain an unapproved food additive.

Despite these regulatory constraints, food manufacturers have experienced considerable success in bringing food additives to market by determining and operating from the assumption that they satisfy the GRAS standard (i.e., "self-affirming GRAS"). In fact, CNN reported that nearly 99 percent of new chemicals used in food and food packaging since 2000 were brought to market through this process, rather than through FDA approval of a petition.⁹

The same article noted that on March 12, 2025, Department of Health and Human Services (HHS) Secretary Robert Kennedy, Jr. publicly announced that HHS intends to eliminate the allowance created by the GRAS program and prevent certain food additives from being allowed to stay on the market under a company's previous GRAS determination. Although FDA may face legal challenges in pursuing this endeavor, food manufacturers should take note of the agency's increased interest in evaluating GRAS determinations and consider reviewing their internal compliance procedures accordingly.

AI could play a valuable role in helping manufacturers identify or develop alternatives to food additives that are on the market based on GRAS determinations. Notably, some food companies are already utilizing AI in product development. For example, Ocean Spray, ADM, and Blue Diamond have worked with a technology company to use its proprietary computational platform that offers an extensive digitized library of natural compounds mapped to human health targets, based on scientific literature and original datasets. The proprietary machine learning platform purportedly allows the company to identify bioactive compounds and predict their effects on human health.¹⁰ As AI technology continues to advance and new platforms emerge, manufacturers may increasingly rely on AI to develop alternatives to food additives at risk of losing GRAS status. The proactive use of AI could help companies anticipate and respond more effectively to evolving food additive regulations.

Existing FDA Guidance on Artificial Intelligence

Although FDA has not yet formalized AI-specific food safety regulations, it has increasingly recognized the potential of AI and machine learning in regulated industries, particularly in the medical device and pharmaceutical sectors. The agency has issued guidance outlining key principles for the responsible use of AI, emphasizing transparency, accountability, reliability, and continuous learning in AI-driven systems.¹¹ For example, in its approach to AI in medical devices, FDA highlights the need for explainability, ensuring that AI models provide clear, interpretable results that can be validated by human oversight.^12,13 These principles are designed to maintain safety and efficacy while allowing AI models to adapt to new data over time.¹³

AI-powered systems utilized in food production and incorporated into food safety plans should similarly aim to be transparent and reliable. A facility's food safety program could try to explain how an AI model detects contamination, predicts risks, or assesses compliance with safety protocols. For instance, an AI-driven supply chain monitoring system that flags potential foodborne pathogen risks should provide clear reasoning behind its alerts, such as detecting patterns in temperature fluctuations or deviations in supplier data.

FDA's approach to AI in other industries highlights the importance of risk management and predetermined change control plans (PCCP), requiring companies to outline how AI models will be monitored, updated, and validated over time.¹³ In food safety, a similar approach could be beneficial in ensuring AI-driven hazard detection systems remain effective as new risks emerge. For example, an AI model designed to detect contaminants in processing environments must continuously refine its algorithms based on new data, but without proper oversight, it could introduce unintended biases or errors. FDA's focus on accountability and structured oversight in medical device AI could serve as a model for developing regulatory frameworks for AI in food safety.

Takeaway

As AI continues to transform the modern commercial landscape, its potential to enhance food safety planning becomes increasingly promising. Real-world applications such as AI-driven supply chain tracking, predictive analytics for contamination risks, and automated monitoring of sanitation practices demonstrate how technology can improve food safety outcomes while enhancing regulatory compliance. AI could also play an important role in helping food manufacturers develop food additive alternatives, which may lead not only to safer products but also a heightened ability to respond to regulatory changes.

However, as AI adoption grows, the need for clear regulatory guidance becomes more pressing. Recent workforce reductions at FDA highlight challenges in keeping pace with evolving technology. While these changes may affect regulatory capacity in the short term, they also present an opportunity for industry stakeholders to take a proactive role in developing responsible AI-driven food safety approaches. Companies that say "welcome to the machine" by integrating AI effectively into their food safety strategies should be better prepared to address future regulatory developments and become more effective in preventing foodborne illness outbreaks.

References

1. Code of Federal Regulations. 21 C.F.R. § 117. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-B/part-117.
2. Qian, C., S.I. Murphy, R.H. Orsi, and M. Weidmann. "How Can AI Help Improve Food Safety?" Annual Review of Food Science and Technology 14 (March 2023): 517–538. https://www.annualreviews.org/content/journals/10.1146/annurev-food-060721-013815.
3. Egbosimba, I. "Change the Future of Food with Automation and AI." Tyson Foods. https://thefeed.blog/2023/11/14/future-of-food-ai-artificial-intelligence-food-industry/.
4. IBM. "Blockchain for Supply Chain Solutions." 2025. https://www.ibm.com/blockchain-supply-chain.
5. United States Code. 21 U.S.C. § 321(s). https://uscode.house.gov/view.xhtml?req=(title:21%20section:321%20edition:prelim).
6. United States Code. 21 U.S.C. § 348(a). https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title21-section348&num=0&edition=prelim.
7. Code of Federal Regulations. 21 C.F.R. § 170.3.
8. U.S. Food and Drug Administration (FDA). "Regulatory Framework for Substances Intended for Use in Human Food or Animal Food on the Basis of the Generally Recognized as Safe (GRAS) Provision of the Federal Food, Drug, and Cosmetic Act: Guidance for Industry." November 2017. https://www.fda.gov/media/109117/download.
9. LaMotte, S. "RFK Jr. Wants to Eliminate FDA's Controversial Food Additive Program. Here's Why That Matters." CNN. March 12, 2025. https://www.cnn.com/2025/03/11/health/gras-reform-kennedy-wellness/index.html.
10. Buss, D. "How AI is Revolutionizing Product Development." Institute of Food Technologists (IFT). October 4, 2024. https://www.ift.org/news-and-publications/food-technology-magazine/issues/2024/october/features/how-ai-is-revolutionizing-product-development.
11. FDA. "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products." Draft Guidance. January 2025. https://www.fda.gov/media/184830/download.
12. FDA. "Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions." December 4, 2024. https://www.fda.gov/media/166704/download.
13. FDA. "Predetermined Change Control Plans for Machine Learning-Enabled Medical Devices: Guiding Principles." Content current as of December 3, 2024. https://www.fda.gov/medical-devices/software-medical-device-samd/predetermined-change-control-plans-machine-learning-enabled-medical-devices-guiding-principles.

Alan Minsk, J.D. is the Chair of Arnall Golden Gregory LLP's Food and Drug practice and Co-Chair of its Life Sciences industry team. He advises pharmaceutical, biologic, medical device, cosmetic and food companies on all legal and regulatory matters relating to the U.S. Food and Drug Administration.

Laura Dona, J.D. is an Associate in Arnall Golden Gregory LLP's Food and Drug practice.

Cody Davis, J.D. is an Associate in Arnall Golden Gregory LLP's Food and Drug practice.

Grace Gluck, J.D. is an Associate in Arnall Golden Gregory LLP's Food and Drug practice.