Halfway to Security and Data Governance for AI Systems in Manufacturing (to be continued…)
One of the main objectives of STAR project is the provision of mechanisms that will ensure the security and data reliability for AI systems in manufacturing. STAR consortium has worked towards achieving this goal, and we are pleased to announce that STAR partners have completed the first version of the AI Security and Data Protection layer of the STAR’s overall architecture.
The figure above illustrates the internal architecture of the AI Security and Data Protection layer of STAR which aims to bridge the gap between the manufacturing plant and the factory security officer by increasing her awareness regarding the cybersecurity posture of the production lines. Several individual components work in synergy like rolling engine gears to convey evidence from the manufacturing environment to the security officer. The goal is to enable informed decision making for mitigation actions and ensure the timely adaptation of the production procedures so that to ensure business continuity and environment’s safety. These gears are:
- AI Cyber Defense for Secure and Trusted AI algorithms: STAR develops AI technologies that secure the operation of the AI systems and algorithms that they comprise. In this direction, the project implements AI Cyber Defense tool that protect and defend AI systems from malicious security attacks. The goal of STAR focuses primarily on defenses against poisoning and evasion attacks against AI-enabled systems. UBITECH has worked for the definition of the architectural design of the AI Cyber Defense tool, as well for the evaluation of state-of-the-art attacks and defenses in the context of the actual STAR pilot environments, leading to the completion of first prototypes system and a scientific publication .
- Runtime Monitoring System (RMS): Enables a real time service that collects security-related data from monitored IoT system components or applications and stores them for further processing. Analytics algorithms analyze the collected data to detect abnormal patterns. Additionally, the collected data feed the logic of the Security Policy Manager which reports incidents exceeding “normal” thresholds. The system is capable of deploying different monitoring probes responsible for the data collection and publishing to the monitoring platform. Netcompany-Intrasoft has already delivered a prototype of the RMS.
- STAR Security Policies Manager (SSPM): Is a tool used to enforce the logic on the detection of abnormal events in a manufacturing environment. The tool has been designed by GFT ITALIA SRL and enables the security/IT officers to configure security policies according to specific business and security requirements. The main purpose of the SSPM is to corelate the evidence collected from the RMS and the AI Cyber defense tool and report the detected cyber security incidents the risk assessment module.
- Risk Assessment and Mitigation Engine (RAME): Complements the SSPM for the visualization of the threats and the corresponding risks. The RAME is based on OLISTIC, UBITECH’s Risk Assessment tool which can support the security officer on getting an overview of the security status of the factory, and more specifically, of the production lines and business processes of interest. Overall, RAME enables the risk management and the identification and visualization of risks through comprehensive and reactive visualization.
- Distributed Ledger Services for Data Reliability (DLSDR): Provides the means for tracking and tracing industrial data for AI algorithms, notably the definitions of the data sources used, the data used to configure STAR AI algorithms and finally the data for persisting their results. To this end, it provides services to the AI algorithms and applications utilizing their results. Netcompany-Intrasoft has provided a prototype of the DLSDR module in order to reinforce the reliability and the security of the source data used in the STAR system by recording information (i.e., metadata) about the acquired data to facilitate the detection of abuse and tampering attempts against these data.
The above-mentioned components comprise the Security and Data Protection layer of the STAR. At the time of writing this blog, the first prototypes of the components have been delivered, meaning that halfway has been covered! The STAR consortium partners look forward to the exiting next steps for the delivery of the full-fledged Security and Data Protection layer of the STAR!
To be continued…..
 Anastasiou, T.; Karagiorgou, S.; Petrou, P.; Papamartzivanos, D.; Giannetsos, T.; Tsirigotaki, G.; Keizer, J. Towards Robustifying Image Classifiers against the Perils of Adversarial Attacks on Artificial Intelligence Systems. Sensors 2022, 22, 6905. https://doi.org/10.3390/s22186905
By: Dimitris Papamartzivanos, UBITECH Ltd