The Data Poisoning Problem in Digital Twin Intelligence

This chapter looks into the underpinnings of the data poisoning threat within digital twin intelligence by analyzing possible attack areas, potential attacker capabilities and how they affect the reliability of twin intelligence. Poisoning manifests primarily in two distinct forms, depending on intended goal: availability attacks, and integrity attacks. Data poisoning is fundamentally different from regular cyberattacks such as Ransomware, which are designed to either steal confidential information or prevent the use of the computer system entirely. The chapter presents probabilistic defenses using uncertainty modeling and statistical inference methods to identify and mitigate poisoned data reported with imperfect information. The survival of an AI-dependent digital twin requires the adoption of probabilistic defense strategies. Upon detection of elevated poisoning risk, the digital twin initiates response actions aimed at containing potential damage while maintaining operational continuity. The chapter proposes a structured framework that provides basis for implementing security mechanisms directly in the cognitive core of digital twins.