Technological Framework for a Secure and Resilient Food Supply Chain

Supply chains in the food industry are part of the critical infrastructure and hence need to be protected against cybercriminals. With digital transformation, the attack surface increases as new digital assets are integrated into the supply chain. Data are collected throughout the supply chain, during production, transport, and storage, and small IoT sensors enable real-time monitoring of goods’ conditions. Manipulations to these sensor data could result in the waste of perfectly good food or in the worst case that customers are sold unpalatable or even unsafe food. In this paper, we propose a three-fold approach to secure the digital supply chain against cyberattacks. First, we advocate the use of end-to-end encryption of the data collected on sensor devices. Despite this being common practice, this is no trivial task, especially when dealing with resource-constrained IoT devices. In order to find suitable security mechanisms and their configurations for different use cases, we propose a model-based resource estimation framework. When data from various sources and parties are integrated, it poses the risk that a party, either unintentionally or deliberately, may introduce contradictory or false information into the system. As a second pillar, this paper presents a trust scoring method that was developed to address this issue, incorporating various data quality and plausibility metrics. Finally, we propose methods to secure the AI model learning pipeline. As artificial intelligence is adopted to critical applications, such as food quality assessment, adversaries can manipulate real-world decisions through attacks. We identify backdoor attacks via data poisoning as a key threat and implement a training-time data separation technique to distinguish poisoned samples from clean data, mitigating backdoor effects and extracting triggers using Explainable AI.