Experts in cybersecurity and artificial intelligence came together recently for a two-day meeting at the National Academies of Sciences, Engineering, and Medicine to examine AI security, how existing cybersecurity tools and practices can be adapted, and emerging risks in high-impact applications such as scientific research, drug discovery, and financial services.

Speakers from various organizations including Microsoft, Meta, Qualcomm, and Security Superintelligence Labs explored topics such as the importance of infrastructure security, threat analysis, frameworks for assessing and benchmarking AI security, and how to characterize and secure the next evolutionary step in generative artificial intelligence — semiautonomous or fully autonomous systems known as agentic AI systems.

“This rapid and pervasive ascent makes the challenge of securing AI systems incredibly urgent,” said Ellen Zegura, senior science and engineering adviser on artificial intelligence at the National Science Foundation, which requested the event. The gathering will inform an issue paper that will outline research priorities and gaps to help coordinate research community efforts around securing AI systems.

Additionally, a separate National Academies rapid expert consultation, Implications of AI for Cybersecurity, is expected in the coming months.