O244-P002-0078 ZTAI: Zero-Trust Artificial Intelligence for secure-by-design O-RAN control and inference

Problem. The Open Radio Access Network (RAN) paradigm as standardized by the O-RAN ALLIANCE is revolutionizing the way cellular networks are designed, deployed, and controlled by promoting AI adoption, open interfaces and interoperability. At the same time, Open RAN can potentially expand the threat surface of cellular systems as (i) once monolithic (and closed) base stations are now disaggregated and interconnected via open interfaces; and (ii) the introduction of RAN Intelligent Controllers (RICs) potentially opens to the execution of voluntarily and involuntarily unsecured xApps/rApps that can affect data confidentiality and integrity, among others. Even though openness and virtualization bring new security challenges and the O-RAN ALLIANCE has introduced a zero-trust architecture (ZTA) to validate and authenticate xApps and other O-RAN applications, our team has shown that even with encryption and ZTA in place, O-RAN systems are vulnerable to adversarial attacks where malicious nodes inject synthetic data, leading AI-based xApps to inefficient control actions and performance degradation. Moreover, our team has shown that the attack can have 100% success rate and always steer away xApps from optimal and effective control actions.