Skip to main content

Designing and Developing Verifiably Safe AI Agents

Francesco Belardinelli ( Imperial College London )

Artificial Intelligence is ubiquitous nowadays. Unfortunately, modern AI systems typically come with no guarantees about their safety, security, compliance, reliability, generalisability. To this aim, Formal Methods are increasingly applied to learning algorithms for their ability to provide strong theoretical guarantees. In this talk I will present some recent works of our lab on Formal Methods in AI about safe reinforcement learning, specifically on shielding RL agents against unsafe behaviours. Our results on this subject are not only of theoretical interest, but have also led to the development of practical safe RL algorithms, which have been implemented in the MASA library, providing a controllable tradeoff between safety and performance.

Speaker bio

Dr Francesco Belardinelli is senior lecturer at the Department of Computing, Imperial College London, where he leads the lab on Formal Methods in AI. He has co-authored more than 100 journal and conference papers on safe AI, the theoretical foundations and applications of Formal Methods to the specification and verification of complex AI systems. On these themes, Dr Belardinelli has led 15 research projects, supported by funding agencies in Italy (SNS), France (ANR), the UK (EPSRC), and the EU. He is deputy director of the CDT in Safe and Trusted AI (director since 2019), for which he has organised the CDT Summer School in 2021, 2023 and 2025. Since 2025 Dr Belardinelli is board member of the European Association for Multi-agent Systems (EURAMAS). He regularly serves in the PC of top conferences (AAAI, IJCAI, AAMAS, ICML, NeurIPS) and acted as workshop chair for AAMAS2021 and IJCAI2026.