arXiv: Privacy, Robustness, and Fairness Trade-offs in Federated Intrusion Detection: Geometric Indistinguishability at the Aggregation Interface

This publication, dated September 2026, is a technical research paper, not a binding regulation. It analyzes the inherent tensions between privacy, robustness, and fairness when applying federated learning to intrusion detection systems. The core finding is that using geometric indistinguishability—a method to obscure data at the aggregation point where model updates are combined—improves privacy but measurably degrades model accuracy and introduces fairness biases across different network traffic profiles. The paper provides a quantitative framework for measuring these trade-offs, which is critical for any organization deploying AI-driven network security.

The affected parties are primarily financial institutions, healthcare providers, critical infrastructure operators, and any EU entity using federated machine learning for cybersecurity threat detection. Under the EU AI Act, such systems may be classified as high-risk if used for network security in critical sectors. This research signals that compliance teams cannot assume privacy-enhancing techniques are automatically compliant; they must verify that these techniques do not introduce discriminatory outcomes or weaken the system's robustness, which are separate legal obligations under the AI Act and GDPR.

Compliance teams should immediately map any federated intrusion detection pilots or production systems against this trade-off framework. Next steps include conducting a differential impact assessment to quantify accuracy loss and fairness disparities caused by privacy layers, documenting these trade-offs in the technical file required for high-risk AI systems, and engaging with the data protection officer to ensure that the chosen privacy mechanism does not violate the principle of data minimisation by degrading security performance. Do not wait for regulatory guidance; proactively test your models for these specific trade-offs and record the results.