aii.news
SecurityBuilders16 Nov

Researchers show malware can hide in a tiny fraction of open-weight AI models

a close up of a computer circuit board
Photo: Klemens Morbe / Unsplash

A 'sleeper agent' technique can hide malicious payloads inside open-weight AI models, using as little as 0.00001% of the neural network. This makes such hidden code very hard for users to detect.

Why it matters

Teams downloading open models should treat them like any third-party software and only use trusted, verified sources.

aii.news picks and summarizes stories from trusted news sources with the help of AI. Every story links to the original reporting. How we work.