Artificial intelligence (AI) has advanced rapidly in recent years, delivering increasingly robust algorithms, more effective solutions, and new applications. Cybersecurity is one of the domains where this progress has been most transformative. While AI-driven cybersecurity has been an active research area for more than two decades - with early applications such as intrusion detection - recent developments have enabled powerful new techniques including deep learning (DL)–based side-channel analysis, evolutionary-computation (EC)–driven security fuzzing, and DL/EC-assisted cryptanalysis.

At the same time, the growing use of machine learning in security-critical systems has raised significant concerns across academia, industry, and society regarding the security, trustworthiness, and privacy of AI systems themselves - from conventional neural networks to modern large language models (LLMs). In addition, emerging research demonstrates that evolutionary computation can be leveraged to automatically generate adversarial prompts and jailbreak strategies that circumvent safety guardrails in large language models.

This tutorial will first provide a concise overview of the domains in which AI, with a focus on DL and EC, is being applied to address cybersecurity problems. We will revisit classical research topics that once dominated the AI-and-security landscape but have received less attention in recent years, placing them in the context of recent developments. The main focus will then shift to emerging research areas that have shown strong momentum and continuous breakthroughs, including fuzzing for vulnerability discovery, implementation attacks, AI-assisted cryptanalysis, and attacks against - and defenses for - AI models.

We will highlight major achievements from the past year and discuss how progress in AI is both accelerating advancements in cybersecurity and reshaping the challenges of securing AI systems themselves. The tutorial will also touch on current regulatory frameworks for responsible and trustworthy AI, and explore the growing intersections between security, large language models, and evolutionary computation. We will conclude by reflecting on how experience in solving AI problems can benefit cybersecurity research and how insights from cybersecurity can strengthen the next generation of AI.