The cyber-AI paradox: Why research must simultaneously advance AI-based defense and secure the AI pipeline

Artificial Intelligence is rapidly transforming cybersecurity, providing new capabilities to detect threats, automate analysis, and respond to increasingly sophisticated attacks. At the same time, the growing adoption of AI introduces a new and complex attack surface: the AI pipeline itself.

This dual challenge—the Cyber-AI paradox—is at the core of the research carried out by GREC, the Cognitive Cybersecurity Research Group, which brings together researchers from i2CAT, UPC, and UOC. The group approaches AI and cybersecurity from two complementary directions: using AI to strengthen cyber defense, while simultaneously ensuring that AI systems and their entire pipelines are secure, trustworthy, and resilient.

During the session, we will present the main research lines currently being explored by GREC and showcase some of the work carried out in these areas, ranging from AI-driven approaches for threat detection and cyber defense to the identification and mitigation of security risks affecting AI models, data, and infrastructures.

The talk will provide an overview of the challenges, ongoing research, and lessons learned at the intersection of AI and cybersecurity, highlighting why advancing AI-based defense and securing the AI pipeline must be addressed together to build the next generation of resilient digital infrastructures.