A Survey on Privacy Attacks Against Digital Twin Systems in AI-Robotics

AI-powered robots and digital twins create new privacy vulnerabilities: The integration of AI/ML with Digital Twin technology in Industry 4.0 robots introduces significant risks for data leakage and model extraction that didn't exist in traditional robotic systems, requiring organizations to rethink their security approaches.

Both AI models and physics-based models are at risk: The survey identifies that attackers can extract not only machine learning models but also first-principles physics-based models from digital twin systems, meaning companies need to protect multiple types of intellectual property and sensitive information embedded in their robotic systems.

Responsible AI practices are essential safeguards: The paper emphasizes that implementing responsible AI frameworks, ethical considerations, and data governance policies are not just compliance measures but active defenses against privacy attacks on AI-robotics systems.

"Trusted autonomy" requires combining multiple disciplines: Rather than treating robotics, AI, and digital twins as separate technical domains, the survey advocates for integrating them with ethical frameworks and trustworthiness principles from the outset to build secure systems from the ground up.

Design choices during model training matter for security: The effectiveness of privacy attacks is directly influenced by decisions made during ML model training and DT integration, suggesting that security and privacy must be considered during development rather than added afterward.