Research

AI for food safety: the FAO report

November 28, 2025

The 2025 FAO report ‘Artificial Intelligence for Food Safety – A Literature Synthesis, Real-World Applications and Regulatory Frameworks’, developed together with researchers from Wageningen University & Research, provides one of the most comprehensive assessments to date of how artificial intelligence is reshaping global food safety governance (van Meer et al., 2025). This exhaustive document meticulously reviews scientific literature, details practical case studies from leading national authorities, and surveys the international regulatory landscape. It positions AI not as a distant future technology, but as a present-day tool with profound implications for enhancing the efficiency, predictive capability, and resilience of food safety systems worldwide, while simultaneously cautioning against its premature or ungoverned use.

Table of Contents

Literature synthesis: mapping AI’s application domains

The report’s foundation is a scoping review of 133 peer-reviewed publications, revealing a steep increase in relevant research since 2012. The analysis categorises AI applications into three primary domains:

the most prominent is ‘scientific advice’, where machine learning (ML) and deep learning (DL) algorithms are revolutionising laboratory processes. These applications range from using convolutional neural networks (CNNs) to identify foodborne pathogens in microscopic images to employing random forests for predicting the virulence of Salmonella strains in ground chicken from genomic data (Karanth et al., 2022; Kang, Park, & Chen, 2020). Furthermore, AI aids in fundamental research, such as modelling the bioaccumulation of heavy metals in soil-crop ecosystems and understanding the environmental factors that lead to microbial contamination (Hu et al., 2020; Toro et al., 2022);

the second domain, ‘inspection and border control’, though less explored, shows significant potential for enhancing risk-based decision-making. Here, models are used to verify food authenticity – such as distinguishing fish species using Fourier transform near-infrared spectroscopy – and to predict which imported food shipments pose the highest safety risk, allowing for targeted sampling and more efficient resource allocation (Liu, Liu, Li, & Wang, 2023; Wu et al., 2023a);

the third domain encompasses ‘activities of competent authorities’, where AI-driven analysis of unstructured data from sources like social media, news reports, and electronic invoices enables the early detection of foodborne illness outbreaks and the identification of emerging consumer concerns (Chen & Zhang, 2022; Sadilek et al., 2018).

Real-world applications: operational innovation in competent authorities

The report presents a diverse set of practical applications already implemented by food safety authorities across Europe, Asia and North America.

In the United Kingdom, regulators have introduced AI-enabled text mining workflows to screen thousands of unstructured documents, extracting weak signals from incident reports, international alerts and scientific literature. These systems support near-real-time surveillance and significantly reduce manual screening efforts while increasing sensitivity to early warning signals.

The Singapore Food Agency (SFA) has deployed automated analytics to monitor global food safety alerts in multiple languages, enabling more targeted import controls for a country heavily dependent on international supply chains. The system applies machine learning to classify event relevance, track trends and identify potential hazards at an early stage.

A particularly advanced application comes from the U.S. Food and Drug Administration (FDA), which has deployed a boosted-tree (LightGBM) machine learning model to predict the probability that an imported food shipment will violate regulatory requirements. By combining data on shipment history, product characteristics, exporting establishment and country risk indicators, the FDA model improves targeting efficiency and increases the likelihood of intercepting unsafe products at the border. The FAO report highlights this system as a leading example of AI-driven import surveillance within a high-volume regulatory environment.

Authorities in Ireland (FSAI) and Italy (Istituto Zooprofilattico Sperimentale, IZS) are also using predictive models, respectively, to identify weak signals that may indicate future food safety risks and to predict whether pathogens have adapted to specific food sources based on genomic data. These models allow more efficient sampling and resource allocation while improving detection rates for microbiological and chemical hazards.

AI is also being used in genomic epidemiology to accelerate outbreak investigations. Machine-learning pipelines can cluster whole-genome sequencing data, identify potential outbreak strains and infer contamination sources more rapidly than traditional analyses. These systems strengthen cross-border collaboration through networks such as INFOSAN, enabling faster responses to foodborne outbreaks.

Lastly, image-based inspection using convolutional neural networks is being trialled for detecting foreign materials, assessing food surface quality and supporting label verification. Early research shows that deep-learning models outperform manual inspection for specific visual tasks when trained on sufficiently large annotated datasets.

Across all these examples, the report stresses three shared principles: the importance of explainability, the role of expert oversight through ‘human-in-the-loop’ systems, and the necessity of clear validation protocols to maintain scientific and regulatory integrity.

The imperative of governance and responsible AI

A core theme of the FAO report is the critical need for robust AI governance and ethical frameworks to ensure trustworthy deployment. The analysis notes a global proliferation of national and international guidelines, such as the European Union’s AI Act and UNESCO’s Recommendation on the Ethics of AI, which recurrently emphasise principles of transparency, fairness, accountability, and human oversight (European Parliament, 2024; UNESCO, 2021).

The report identifies key challenges that must be managed, including:

the ‘black box’ problem of complex algorithms, where explainable AI (XAI) techniques like SHapley Additive exPlanations (SHAP) are necessary to elucidate model decisions (Lundberg & Lee, 2017);

data bias, which can perpetuate existing inequalities if training data is unrepresentative. The data on which AI systems are trained can significantly and unexpectedly change over time, affecting system functionality and trustworthiness (NIST, 2023);

the potential for AI ‘hallucinations’, where models generate plausible but fabricated information;

premature use of artificial intelligence. ‘The risk of prematurely using AI in food safety, whether by applying techniques that are not yet suitable for the specific data or problem or by implementing AI without the necessary expertise to interpret its output, lies in potentially undermining the trust and credibility of the organization employing it’ (Santoni de Sio and Mecacci, 2021; Smith, 2018).

Underpinning all effective AI systems is the principle of data governance. The report strongly advocates for the adoption of the FAIR principles – making data Findable, Accessible, Interoperable, and Reusable – as a foundational prerequisite for developing accurate and reliable models (Wilkinson et al., 2016). For countries, particularly low- and middle-income countries (LMICs), facing significant data gaps, the report advises prerequisite activities such as building data collection strategies, developing national AI policies, and investing in capacity development, rather than rushing into AI development prematurely.

Interim conclusions

The FAO report concludes that AI holds transformative potential for creating more proactive, efficient, and robust food safety systems. However, its successful integration is a strategic endeavour, not a mere technical upgrade. For competent authorities considering adoption, the report offers clear guidance: first, identify a well-defined problem that AI can solve; second, rigorously assess the availability and quality of required data; and third, invest in building AI literacy and cross-disciplinary collaboration among food scientists, data analysts, and policymakers. By prioritising responsible governance, equitable access, and human-centric design, the global community can harness artificial intelligence to build safer agrifood systems for all, ensuring that technological advancement translates into tangible public health benefits.

Dario Dongo

References

Chen, Y., & Zhang, Z. (2022). Exploring public perceptions on alternative meat in China from social media data using transfer learning method. Food Quality and Preference, 98, 104530. https://doi.org/10.1016/j.foodqual.2022.104530

European Parliament and the Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). https://eur-lex.europa.eu/eli/reg/2024/1689/oj

Hu, B., Xue, J., Zhou, Y., Shao, S., Fu, Z., Li, Y., Chen, S., Qi, L., & Shi, Z. (2020). Modelling bioaccumulation of heavy metals in soil-crop ecosystems and identifying its controlling factors using machine learning. Environmental Pollution, 262, 114308. https://doi.org/10.1016/j.envpol.2020.114308

Kang, R., Park, B., & Chen, K. (2020). Identifying non-O157 Shiga toxin-producing Escherichia coli (STEC) using deep learning methods with hyperspectral microscope images. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 224, 117386. https://doi.org/10.1016/j.saa.2019.117386

Karanth, S., Tanui, C. K., Meng, J., & Pradhan, A. K. (2022). Exploring the predictive capability of advanced machine learning in identifying severe disease phenotype in Salmonella enterica. Food Research International, 151, 110817. https://doi.org/10.1016/j.foodres.2021.110817

Liu, H., Liu, H., Li, J., & Wang, Y. (2023). Rapid and accurate authentication of Porcini Mushroom species using Fourier transform near-infrared spectra combined with machine learning and chemometrics. ACS Omega, 8(22), 20095–20106. https://doi.org/10.1021/acsomega.3c01229

Lundberg, Scott & Lee, Su-In. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30. https://doi.org/10.48550/arXiv.1705.07874

NIST (National Institute of Standards and Technology, US Department of Commerce). 2023. Artificial Intelligence Risk Management Framework (AI RMF 1.0). https://doi.org/10.6028/NIST.AI.100-1

Sadilek, A., Caty, S., DiPrete, L., Mansour, R., Schenk, T., Bergtholdt, M., Jha, A., Ramaswami, P., & Gabrilovich, E. (2018). Machine-learned epidemiology: real-time detection of foodborne illness at scale. NPJ Digital Medicine, 1(1), 36. https://doi.org/10.1038/s41746-018-0045-1

Santoni de Sio, F. and Mecacci, G. (2021). Four responsibility gaps with artificial intelligence: Why they matter and how to address them. Philosophy and Technology, 34(4), 1057-1084. https://doi.org/10.1007/s13347-021-00450-x

Smith, M.J. (2018). Getting value from artificial intelligence in agriculture. Animal Production Science, 60(1), pp.46-54. https://doi.org/10.1071/AN18522

Toro, M., Weller, D. L., Ramos, R., Diaz, L., Álvarez, F. P., Reyes-Jara, A., Moreno-Switt, A. I., Meng, J., & Adell, A. D. (2022). Environmental and anthropogenic factors associated with the likelihood of detecting Salmonella in agricultural watersheds. Environmental Pollution, 306, 119298. https://doi.org/10.1016/j.envpol.2022.119298

UNESCO. (2021). Recommendation on the Ethics of Artificial Intelligence. https://unesdoc.unesco.org/ark:/48223/pf0000380455

van Meer, F., van der Velden, B., & Takeuchi, M. (2025). Artificial Intelligence for food safety – A literature synthesis, real-world applications and regulatory frameworks. FAO. https://doi.org/10.4060/cd7242en

Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., Appleton, G., Axton, M., Baak, A., Blomberg, N., Boiten, J. W., da Silva Santos, L. B., Bourne, P. E., & Bouwman, J. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data, 3(1), 160018. https://doi.org/10.1038/sdata.2016.18

Wu, L.-Y., Liu, F.-M., Weng, S.-S., & Lin, W.-C. (2023a). EL V.2 Model for Predicting Food Safety Risks at Taiwan Border Using the Voting-Based Ensemble Method. Foods, 12(11), 2118. https://doi.org/10.3390/foods12112118