Detecting model weight tampering in GPU memory (Rome attacks)
AI summary
Editorial policyThe article explains Rank-One Model Editing (ROME) attacks that corrupt model weights in GPU memory, why CPU-focused tools miss them, and how Stealthium detects such tampering.
Key points
- •It introduces ROME attacks, where a malicious Python module rewrites model weights in GPU memory to corrupt answers.
- •It explains why CPU-focused security tooling fails to detect this class of attack.
- •It describes how Stealthium's runtime sensor detects model tampering by tying CUDA kernel launches back to the issuing Python code.
#Security & Misuse#Semiconductors & Infrastructure#Stealthium