Securing AI Infrastructure: Lessons from National Cybersecurity Strategies and Attacks Against Other Critical Sectors

The talk argues that AI labs are attractive espionage targets and that the AI sector is not yet treated as a dedicated critical infrastructure sector in the US. The speakers draw on a study of more than 200 cyber attacks, insider threats and espionage cases across industries such as semiconductors and biotech, tracking what happened to victims and their wider sectors afterward.

The findings they described include lasting revenue drops for breached firms, damage that spreads across an entire sector, a shift from trade secrets toward patents, and a sharp decline in Asian research staff after incidents. That last point suggests a need for response plans that avoid panic-driven responses. They also reviewed national cyber strategies from countries including the US, Australia, Singapore and the UK. AI is often covered in these documents, but securing it receives less attention than AI safety, and metrics, incentives and clear accountability are weak.

Recommendations included defining concrete security requirements for AI as critical infrastructure, separating incident reporting from law enforcement as Australia has done, and using UK-style programs in which industry staff hold pre-approved clearances while keeping their jobs. They also called for wider security clearance coverage and more non-technical cyber expertise in law and policy. The session closed with audience questions on using supply chain leverage and on incentive structures.