Digital Security and the Disruptive Nature of Artificial Intelligence

Artificial Intelligence has a rich scientific history spanning over five decades. Recent breakthroughs in generative AI, driven by Large Language Models (LLMs), surprised the general public with astonishing applications that demonstrate near-human performance and are believed to reshape not only the digital economy but society at large.

This presentation delves into a security-centric analysis of Machine Learning systems, particularly focusing on Generative AI like ChatGPT. We explore both the threats posed to these systems and the potential threats that emanate from their deployment. Additionally, we examine the utilization of such systems in security applications, highlighting our recent work on BGP traffic prediction as a practical example.

The discussion extends to the transformative impact of Generative AI on building secure and reliable systems. By introducing novel attack vectors, these systems present significant challenges to the ongoing digital evolution of our societies. We contend that the properties inherent in these systems demand urgent attention to ensure security and reliability, emphasizing the potential for disruption unless research in these domains keeps pace with the real-world deployment of AI components. As we stand at the crossroads of AI-driven transformation, understanding and addressing the security implications are paramount to realizing the full potential of this technology.