Get the latest news and updates on FINTECH.TV

AI’s Biggest Security Blind Spot Is Its Infrastructure

AI investment is expected to top $1 trillion this year, but new research may expose a blind spot in the boom. On Taking Stock, JD Durkin speaks with Yuriy Bulygin, founder and CEO of Eclypsium, about the security risk in the infrastructure that powers AI.

Bulygin says much of the cybersecurity conversation around AI focuses on the upper layers, such as protecting applications, autonomous agents and data from prompt injection, which he calls legitimate and important. He says all AI workloads and models run on massive infrastructure, and that there is no trusted or secure AI without trusted infrastructure underneath it. He warns that if attackers compromise that infrastructure, they can steal data and modify the outputs and behavior of the models and agents people rely on in their personal lives, in enterprises and in national security.

Describing Eclypsium’s new research, Bulygin says determined adversaries, including nation-states, can take over the infrastructure that models and agents run on and control the operating environment without touching the model itself. He says that could let attackers push agents to build backdoors or vulnerabilities into generated code, or change a model’s behavior in real time to give a wrong and potentially dangerous answer to a health question.

Advertisement

Latest articles

Related articles