GMLaaS

New attack weaponizes GNN explainability for model extraction

Researchers have developed a new method to extract information from graph neural networks (GNNs) by exploiting their explainability interfaces. This attack, operating under strict black-box constraints, uses explanation…