"ShadowLogic Attack Targets AI Model Graphs to Create Codeless Backdoors"

According to the Artificial Intelligence (AI) security company HiddenLayer, codeless, persistent backdoors can be planted in Machine Learning (ML) models by manipulating an AI model's graph. The "ShadowLogic" technique manipulates a model architecture's computational graph representation to initiate behavior defined by the attacker in downstream applications, thus enabling AI supply chain attacks. HiddenLayer notes that threat actors can implant codeless backdoors in ML models using ShadowLogic that persist through fine-tuning and can be used in highly targeted attacks. This article continues to discuss findings regarding the ShadowLogic attack.

SecurityWeek reports "ShadowLogic Attack Targets AI Model Graphs to Create Codeless Backdoors"

Submitted by grigby1