AI/ML, Threat Intelligence

NVIDIA GPU-run AI models at risk of novel GPUHammer attacks

(JHVEPhoto – stock.adobe.com)

Artificial intelligence models run with NVIDIA GPUs could have their accuracy decrease from 80% to 0.1% through the new RowHammer attack variant dubbed "GPUHammer", reports The Hacker News.

With a significant reduction in AI model accuracy resulting from the tampering of ImageNet deep neural network models with a single-bit flip, GPUHammer could also be exploited to facilitate model parameter corruption, a study from University of Toronto researchers revealed. "Enabling Error Correction Codes (ECC) can mitigate this risk, but ECC can introduce up to a 10% slowdown for [machine learning] inference workloads on an A6000 GPU," said researchers. NVIDIA has likewise recommended the activation of system-level ECC to protect GPUs from the intrusion. "Risk of successful exploitation from RowHammer attacks varies based on DRAM device, platform, design specification, and system settings," noted NVIDIA. Such a development comes after NTT Social Informatics Laboratories and CentraleSupelec researchers' discovery of another RowHammer attack derivative dubbed "CrowHammer," which could compromise the Falcon post-quantum signature scheme.

An In-Depth Guide to AI

Get essential knowledge and practical strategies to use AI to better your security program.

Get daily email updates

SC Media's daily must-read of the most current and pressing daily news

By clicking the Subscribe button below, you agree to SC Media Terms of Use and Privacy Policy.

You can skip this ad in 5 seconds