UAAD: A unified anomaly-aware defense against diverse backdoor attacks on graph neural networks

Summary

UAAD (Unified Anomaly-Aware Defense) is a framework that mitigates diverse backdoor attacks on graph neural networks. It builds a pseudo-clean reference distribution with spectral clustering, scores suspicious nodes with Mahalanobis distance, and prunes suspicious edges with an adaptive Gaussian Mixture Model, without fixed thresholds. In GCN-based evaluation on five real-world datasets, its mean ASR stays at or below 0.0775 against GTA, UGBA, and DPGBA, with a dataset-dependent clean-accuracy trade-off.

Mitigation

UAAD is the proposed defense: an anomaly-aware framework that removes suspicious edges through adaptive GMM-based pruning and uses edge-level anomaly priors for joint training.

Topics

• Adversarial machine learning

Related items

• Medium'Aokigahara' Stealer Targets 'Developer Keyrings' & AI AssistantsSimilar attack · Dark Reading
• HighGHSA-3w55-933p-8pv5: Strands Agents: Local Privilege Escalation via Symlink Attack in Default Session File StorageSimilar attack · GitHub Advisory Database
• HighGHSA-g5ff-637f-6q2m: LiteLLM: internal_user_viewer arbitrary local file read via `vertex_ai_credentials`Similar attack · GitHub Advisory Database
• MediumGHSA-hhww-mrg2-969h: LiteLLM vulnerable to reflected XSS in /sso/debug/callbackSimilar attack · GitHub Advisory Database
• HighCVE-2026-117573: hagent OS command injection through args in arduino and esp32 MCP toolsSimilar attack · NVD/CVE Database