As artificial intelligence systems become increasingly autonomous and integrated into software engineering (SE) workflows, they introduce novel cybersecurity risks that are not yet well understood. This project investigates the emerging security challenges posed by agentic AI—systems based on large language models (LLMs) that can retain memory, use external tools, and modify software artifacts across multiple steps. These capabilities make such agents both valuable and vulnerable, particularly in high-impact development tasks like code generation, patching, testing, or repository management.

This bilateral project aims to map the security risks of agentic AI in SE tasks and develop practical, cost-sensitive defenses. Through collaborative research, we will explore how AI agents can be exploited or manipulated, and assess lightweight mechanisms for securing their operation. The Austrian team (TU Wien) will focus on attack surface analysis and agent misuse simulation, while the Polish team (Nicolaus Copernicus University) will lead the development and evaluation of defenses based on software repository mining and empirical analysis.