Food safety is a critical public health concern, with foodborne diseases affecting millions globally. The growing complexity of food additive regulations across different countries makes it challenging to streamline decision-making processes for ensuring safety. Additionally, data is often complex and difficult to interpret, and discrepancies among regulatory frameworks make comparative analysis both time-consuming and costly. Large Language Models (LLMs) like ChatGPT-4 offer potential solutions by generating explanations and reasoning about whether food additives should be banned or allowed. This paper evaluates the reasoning capabilities of LLMs in food safety by analyzing a dataset of food additives, comparing the model’s predictions with human assessments. The results indicate that while the model can provide logical justifications for many decisions, there are significant limitations. These include issues with data accessibility, hallucinations, and cultural discrepancies in regulations that affect accuracy. We provide a detailed analysis of these challenges and discuss how LLMs can be refined for reliable integration into real-world food safety assessments. Our findings highlight both the strengths and limitations of using Artificial Intelligence in regulatory environments and suggest avenues for future research and development.