UBITECH announces the launch of the cPAID (Cloud-based Platform-agnostic Adversarial AI Defence) Research and Innovation Action, a cutting-edge European Commission-funded project (Grant Agreement No. 101168407). The project officially commenced on October 1, 2024, and will run until September 2027. The kick-off meeting took place in Athens, Greece, on October 22-23, 2024, gathering key stakeholders to align on the vision and strategic objectives of the initiative. The cPAID project aims to research, design, and develop an advanced defense framework that provides comprehensive protection for AI applications and operations from malicious actions and adversarial attacks. This pioneering cloud-based, platform-agnostic solution will address critical vulnerabilities in AI systems by combating both poisoning and evasion adversarial attacks.
cPAID will leverage state-of-the-art techniques such as lifelong semi-supervised reinforcement learning, transfer learning, feature reduction, adversarial training, and privacy-preserving technologies. The framework will also incorporate explainable AI (XAI), generative AI, context-awareness, and risk assessment tools to offer organizations robust defenses. The ultimate goal is to ensure the security, privacy, and ethical excellence of AI systems, aligning with EU principles for AI ethics. The project also seeks to establish guidelines for: Integrating security and privacy-by-design in AI application development; Assessing and enhancing the robustness and resilience of machine learning and deep learning algorithms; Ensuring adherence to ethical standards in AI systems; and Validating AI performance through real-life use case scenarios. These efforts will further drive research toward developing certification schemes that can verify the robustness, security, privacy, and ethical performance of AI applications.
As a key partner, UBITECH plays a pivotal role in the integration and validation of the cPAID platform. UBITECH will contribute in the development of the cPAID adversarial AI attack cyber range, an innovative training environment for AI security professionals. This cyber range will simulate adversarial AI attacks on AI models, enabling organizations to conduct advanced training and security testing. The cyber range will include simulations of vulnerable AI systems, Red Team-Blue Team exercises, AI-Enhanced Social Engineering, and AI in IoT security scenarios. Built on open-source game engines and interactive storytelling frameworks, the range will support both individual and large-scale training exercises with a scoring mechanism based on Capture-The-Flag (CTF) methodology. This comprehensive platform will follow established frameworks like the Serious Games Design Framework and the Gamification Design Framework, providing an immersive and educational experience for cybersecurity practitioners.
