Need to Know

• AI's potential food safety applications include pathogen and chemical hazard detection, food fraud and authenticity, outbreak surveillance, risk forecasting, and regulatory decision-making
• Microbiological hazard detection represented the largest area of research, followed by chemical hazards
• Barriers to wider adoption remain, such as data quantity, quality, and accessibility; confidentiality; and transparency.

AI Applications for Microbiological and Chemical Hazards

Microbiological hazards represented the largest research area identified by the review, accounting for 56 of the 161 included papers, or 35 percent. Of those studies, 33 (59 percent) applied AI to improve the accuracy, speed, cost, or efficiency of laboratory testing. Applications included the detection or quantification of pathogens such as Salmonella, Escherichia coli, and Listeria using data generated through whole genome sequencing (WGS), microscopy, molecular methods, colony counts, and optical spectroscopy.

Another 11 microbiological hazard studies examined AI for forecasting future risks, while 12 investigated risk factors or pathogen etiology. For example, one study used machine learning and Salmonella genome sequences from ground chicken to predict disease outcomes, while another found that physicochemical water quality and weather data combined with random forest algorithms could be used to predict E. coli levels in agricultural water.

Chemical hazards were addressed by 40 studies (25 percent). Half focused on improving laboratory testing, including applications targeting pesticides, veterinary drug residues, mycotoxins, and heavy metals. The researchers noted a trend toward pairing machine learning with onsite, nondestructive, lower-cost instrumentation intended to approximate results from more expensive and labor-intensive laboratory analytical methods.

Food Fraud, Outbreak Surveillance, and Regulatory Applications

A total of 27 studies (17 percent) examined AI for food fraud or authenticity. Most used analytical technologies such as near-infrared and fluorescence spectroscopy, electronic noses, and electronic tongues. Some research also explored regulatory applications, including an automated alarm system that analyzed nearly 100 million electronic invoices to identify potentially problematic edible oil manufacturers.

Additionally, nine studies focused on foodborne disease outbreaks and surveillance. These included applications using combined datasets to identify outbreak risk factors, forecast outbreaks, and improve inspection prioritization. One study analyzed anonymized internet search and smartphone location data to identify potential sources of foodborne illness in real time, with the approach improving identification of potentially problematic establishments by more than threefold compared with traditional methods.

Another 29 studies addressed broader food safety applications, including early warning systems, food spoilage, inspection reports, recalls, and border inspections. In one example, a machine learning-based border inspection model flagged imported food cases with noncompliance rates up to three times greater than those identified through random sampling.

Data Quality, Transparency Present Challenges for Food Safety AI

Despite the range of potential applications, the researchers identified significant barriers to wider and more reliable use of AI in food safety.

Food safety datasets are often incomplete or highly imbalanced because most analytical samples have low contamination levels, while relatively few exceed maximum limits. Such imbalances can result in models that accurately identify low-risk cases but perform poorly at recognizing the comparatively rare, higher-risk cases that are particularly important to food safety decision-making.

The authors also highlighted missing metadata and limited data accessibility as obstacles. Information such as sample origins, sampling locations, and sampling conditions can be essential for tracing contamination, identifying risk factors, and forecasting hazards. The researchers called for standardized data collection consistent with FAIR principles, meaning data should be Findable, Accessible, Interoperable, and Reusable.

At the same time, food safety data can contain private, commercially sensitive, or regulatory information that makes open sharing difficult. The researchers identified federated learning, in which models interact with decentralized datasets without requiring the underlying sensitive information to be centralized, as one potential approach to overcoming this challenge.

Transparency is another concern, particularly for deep learning models whose decision-making can be difficult to interpret. The researchers pointed to explainable AI (XAI), which seeks to identify the parameters driving AI-generated decisions, as an increasingly important tool for detecting bias and supporting transparency and accountability in food safety applications.

Data Lacking on the Use of AI by Food Producers

The review also identified a notable gap in the scientific literature: relatively few peer-reviewed studies examined AI applications within food production, processing, and manufacturing businesses. The authors suggested this could partly reflect industry's use of whitepapers rather than scientific journals to communicate findings, as well as proprietary and commercial considerations that discourage publication of AI-related work.

The lack of data on industrial AI use may not imply that AI is not being applied in these settings, however. In the “Food Safety Insights” column from the June/July 2026 issue of Food Safety Magazine, an industry survey revealed that AI has already moved into food safety programs; sometimes quietly, even when companies insist AI not be used.

Future Needs and Moving Forward

The study authors concluded that AI holds substantial potential for food safety, but its reliability and practical value will depend on overcoming challenges involving data quality and availability, privacy, governance, and model interpretability. FAO has also previously stressed the importance of robust governance and responsible AI adoption for food safety.

WFSR has announced plans to organize the first AI for Food Safety (AIFS) conference in 2027 with international AI and food safety researchers, providing a forum to discuss emerging applications, research, and challenges associated with the technology.