Dr Francesca Boem, University College London

The challenge  

Critical infrastructure systems – such as electricity grids, water networks, and healthcare systems – are undergoing a massive transformation driven by the Internet of Things and AI. This convergence of IT and operational technology brings exciting opportunities to use data for real-time monitoring, distributed control, and autonomous optimisation to make systems more efficient and sustainable. However, can AI be reliably deployed for the control of safety-critical systems? What happens if the information AI uses is not accurate? Even worse, what if attackers intentionally modify the data acquired by the AI in real time, modifying the system's behaviour without being detected and causing catastrophic physical, economic, or societal harm? Attacks specifically targeting AI-driven control applications have not been sufficiently investigated, despite their potentially severe consequences. Developing frameworks to guarantee the safety and reliability of these interconnected architectures is an imperative.