Unified privacy and defense mechanisms

Establish unified frameworks that combine privacy mechanisms with defenses against Byzantine and other malicious agents while preserving sufficient signals to detect graph-structure and embedding attacks.

Background

Current anonymization methods primarily protect against honest-but-curious participants, whereas malicious agents may manipulate graph structures or node features. Privacy mechanisms can conceal the signals needed for detecting such attacks, creating a tension between privacy and integrity protection. The survey identifies the combination of these defenses as unresolved.

References

However, how privacy mechanisms can be combined with additional defenses (e.g., embedding filtering~\citep{he2024privacy}) against sophisticated adversaries remains unclear, since privacy guarantees often hide the very signals needed for detecting these adversarial threats, calling for unified frameworks that jointly address honest-but-curious and malicious threat models.

— From Euclidean to Graph-Structured Data: A Survey of Collaborative Learning  (2609.02984 - Bourgerie et al., 2 Sep 2026) in Section 7, paragraph “Open challenges”

Graph perturbation and structural anonymization could reduce the discriminative power of the features driving the attack, while differential privacy offers formal guarantees, though its adaptation to heterogeneous KGs remains an open problem.

— Link Inference Attack on Privacy-Preserving Knowledge Graphs  (2609.39362 - Bouguerra et al., 30 Sep 2026) in Section Conclusion {content} Future Work

How can federated agent systems protect privacy while remaining secure against unreliable or malicious contributions?

— Federated Agent Optimization  (2610.01195 - Yang et al., 1 Oct 2026) in Section 7, Future Works, item 183