AOAC Committee on Safety & Security – “Safety & Security Pills”
A concise safety and security reflection on AI reliability, laboratory risk, and the need for human oversight, prepared for the AOAC Committee on Safety & Security by Prof. Dr. Salvatore Parisi
In-the-Lab Safety and Security: Artificial Intelligence and Reliable Evaluation
High-level summary: AI can support laboratory safety by predicting risks and guiding decisions, but reliability, transparency, privacy, resistance to manipulation, and human oversight are essential before AI/ML tools are used in chemical and microbiological laboratories.
Background
Artificial Intelligence (AI) is now one of the most active topics in technology and is increasingly present in daily decision-making. AI tools appear ready to provide advice, support, and succinct answers to both simple and complex questions. However, an important safety question remains: are we using AI correctly?
This question is especially relevant in scientific research, media activity, and chemical and microbiological laboratories, where AI-supported outputs may influence safety and security decisions. AI may help predict laboratory accidents and suggest safer behaviours, but only when its answers are reliable, interpretable, and appropriately supervised.
Reliability as a Safety Requirement
Reliability may be defined as the ability of AI systems to function consistently across different contexts when parameters, data, or user behaviours vary (Verma et al. 2016). In practical terms, AI reliability can be evaluated by examining the frequency and nature of errors or failures. Tools such as the “bathtub curve,” which represents failure rate over time, can help assess variability and robustness in AI systems (Nguyen et al. 2024).
Questions to Ask Before Using AI/ML Tools
AI systems often reach users through machine learning (ML) interfaces. Because the reliability of ML-related answers depends on the reliability of the underlying AI system, the following questions should be considered before relying on AI/ML tools without careful human supervision (Nguyen et al. 2024):
- Can ML and AI systems be interpreted by humans, including clear reasons for their decisions, or do they function as “black boxes”?
- Are AI systems designed to resist intentional manipulation?
- Can human users intervene in the decision-making process when needed?
- Do ML and AI systems adequately protect data privacy?
- Can ML systems communicate uncertainty levels or confidence estimates for their outputs?
- Are AI/ML development and training responsibilities appropriately separated and controlled?
Implications for Laboratory Safety and Security
These questions are important across many fields, but they are particularly critical in chemical and microbiological laboratories, where safety precautions, security practices, accident prediction, and disaster prevention are central responsibilities. AI can provide powerful support, but careful management, validation, and human evaluation remain essential. The AOAC Committee on Safety and Security continues to engage in this and other areas within the broader safety and security landscape (Parisi 2024; Parisi 2025).
Takeaway for posting: AI should be treated as a decision-support tool, not a replacement for trained scientific judgment, validated procedures, and accountable human oversight.
References
Nguyen TH, Saghir A, Tran KD, Nguyen DH, Luong NA, Tran KP (2024). Safety and reliability of artificial intelligence systems. In Artificial intelligence for safety and reliability engineering: Methods, applications, and challenges (pp. 185-199). Cham: Springer Nature Switzerland
Parisi S (2023) Awareness of Safety and Security in the Laboratory. New Perspectives. Inside Laboratory Management 27, 6:7-8
Parisi S (2025) AI-based Testing Methods and Food Allergens Detection against Labelling Errors and Frauds. Proceedings of the First International Conference of Artificial Intelligence in Health Science, Amman As Salt, Jordan, 29-30th December 2025
Verma A, Srividya A, Karanki DR (2016) Reliability and Safety Engineering: Second Edition. 01 2016. ISBN 978-1-4471- 6268-1
Photo by Ai HUB – stock.adobe.com
