Beyond the CPU: Exposing AI Attacks in the GPU Stack
Summary
This article discusses the security blind spots in AI infrastructure, which largely runs on GPUs while most security tools focus on CPUs. It highlights how attackers can exploit the GPU stack, bypassing traditional security measures. The session will cover the GPU attack surface, identify missed threats by current tools, and provide strategies for defenders to close this visibility gap.
IFF Assessment
The article details how attackers can exploit weaknesses in GPU-based AI infrastructure, creating new attack vectors that traditional security tools may miss.
Defender Context
Defenders need to be aware that traditional CPU-centric security tools like EDR and SIEM may not provide adequate visibility into GPU-based AI infrastructure. Attackers are increasingly targeting the GPU stack, utilizing novel techniques that leave minimal traces on the CPU, necessitating a shift in security monitoring and tooling to encompass GPU-specific telemetry and attack patterns.