Mmastodon BusinessStartups first seen 5 d ago, last 5 d ago, peak #3

New method detects AI model tampering in GPU memory

Original: Detecting model weight tampering in GPU memory (Rome attacks) Article URL: https:// stealthium.io/blog/hands-off-d etect

Stealthium has published a technical post describing how to detect tampering with AI model weights directly in GPU memory, focusing on so-called Rome attacks, where malicious modifications are made to model weights at runtime. The technique aims to spot compromised models without direct access, a growing concern as AI systems are increasingly targeted by supply-chain style attacks.

Why now: Security researchers and AI practitioners are increasingly worried about runtime attacks on deployed models, making new detection techniques notable

StealthiumRome attacksGPU memory

Open on mastodon →

Evidence

• Detecting model weight tampering in GPU memory (Rome attacks) Article URL: https:// stealthium.io/blog/hands-off-d etecting-model-weight-tampering Comments URL: https:// news.ycombinator.com/item?id=5 0152976 Points: 5 # Comments: 0 https:// stealthium.io/blog/hands-off-d… · techwire@gamefan.net · 1

API: https://socialmediatrends-api.osmike.com/v1/trends/2957754