New attacks reveal significant privacy risks in Graph Neural Networks

Researchers have developed two new generative reconstruction attacks, the graph-label conditioned (GLC) attack and the embedding-label conditioned (ELC) attack, to probe the privacy vulnerabilities of Graph Neural Networks (GNNs). These attacks leverage target model predictions and intermediate representations to reconstruct sensitive graph data, demonstrating that adversaries can generate high-quality graphs in black-box scenarios. The study also introduced a variant with reduced query requirements that maintains strong performance, highlighting GNNs' susceptibility to privacy breaches across various noise scales. AI

IMPACT Highlights potential privacy risks in AI models used for graph data analysis, necessitating further research into GNN security.

RANK_REASON The cluster contains an academic paper detailing novel research on privacy attacks against Graph Neural Networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• AIDS
• ELC attack
• GLC attack
• Graph Neural Network Models
• NCI1
• PROTEINS

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →