As one of Italy’s leading research institutions, CNIT (Consorzio Nazionale Interuniversitario per le Telecomunicazioni) plays a key role in AI-driven cybersecurity research, bringing expertise in adversarial AI attack generation, complexity and cryptography, and AI-based defense mechanisms. Within AIAS, CNIT leads the development of AI-powered attack tools, security assessments, and cybersecurity innovations to enhance AI security.
Key Contributions:
🔹 Semi-Supervised Learning Model & AI Attack Weaponizer
CNIT is responsible for designing and implementing an advanced AI-based adversarial attack engine, which:
- Utilizes semi-supervised learning and attack graphs to automatically generate adversarial AI attacks.
- Develops an API that acts as a proxy between AI attack generators and target AI systems, allowing real-time adversarial testing and security assessments.
- Enhances AIAS’s capabilities in understanding and mitigating AI-specific cyber threats.
🔹 AI Attack Taxonomy & Threat Detection
CNIT contributes to the definition and classification of AI threats, helping AIAS develop a comprehensive framework for AI-based attack detection and mitigation strategies. Additionally, CNIT’s role includes:
- Analyzing cybersecurity data to extract meaningful insights and mitigation strategies.
- Developing AI-driven adversarial attack detection mechanisms to safeguard AI models from sophisticated cyber threats.
- Leveraging expertise in OS-level security and malware analysis to enhance the platform’s defense capabilities.
🔹 System Integration & AIAS Platform Development
CNIT plays a crucial role in integrating AIAS modules and ensuring seamless communication between different system components. Their private cloud infrastructure and 4-5G testbed provide a robust environment for evaluating AIAS functionalities.
🔹 Cybersecurity Innovation & AI Attack Simulation
CNIT contributes by:
- Developing a tool for executing existing and 0-day cyberattacks on AI-based systems, allowing for robust security assessments.
- Leveraging PEzoNG, an advanced evasion technique for Windows systems, and Whisper2Shout, a novel unhooking technique, to improve AIAS’s AI attack detection and response capabilities.
- Designing an AI monitoring agent that extracts valuable insights from malware execution, supporting the AIAS adversarial AI defense framework.
🔹 Collaboration & Knowledge Transfer
Through secondments at PDM and BEIA, CNIT enhances its expertise in adversarial AI attack generation, detection, and cybersecurity tools. These collaborations strengthen CNIT’s ability to:
- Understand business aspects of adversarial AI threats.
- Gain hands-on experience with AI security tools and prototype development.
- Publish high-quality research papers and contribute to European cybersecurity initiatives.
🔹 Workshops & Industry Engagement
CNIT and PDM will co-organize the fourth AIAS workshop, fostering discussions on AI security and advancing research in cybersecurity attack automation, deception mechanisms, and AI adversarial resilience.
With its deep expertise in adversarial AI, OS-level security, and AI-powered attack detection, CNIT plays a crucial role in strengthening AIAS’s cybersecurity framework. By combining cutting-edge AI attack simulations, innovative detection mechanisms, and real-world cybersecurity applications, CNIT ensures that AIAS remains at the forefront of AI security research.
Stay tuned as CNIT and AIAS continue pioneering the future of AI-driven cybersecurity!