Secure Software Development Processes Based on Agentic Systems

The project aims to define and validate secure software development processes supported by AIbased agentic systems. The agents will operate throughout the development lifecycle to analyse requirements, generate and review code, produce tests, and detect vulnerabilities, technical debt, and quality issues. The project will combine language models trained on code with static analysis and software repository mining. Particular attention will be paid to both the security of the resulting software and the reliability of the agentic process. Controls, verification criteria, and oversight mechanisms will be defined to reduce the risk of agents introducing vulnerabilities, propagating errors, or adopting insecure practices. The models’ ability to generalise across languages and contexts will also be examined. The results will include verifiable and reproducible processes, datasets, models, tools, and guidelines for the responsible use of agents...