Proactive and privacy-Preserving defense for DNS over HTTPS via federated AI attestation (PAFA-DoH)

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Cryptography

Background:

• DNS over HTTPS (DoH) offers privacy but remains vulnerable to tunneling, resolver manipulation, and side-channel leakage.
• Existing defenses struggle to address sophisticated threats in encrypted DNS traffic.

Purpose of the Study:

• To introduce PAFA-DoH, a unified framework for privacy-preserving, post-quantum hardened protection of encrypted DNS.
• To develop a system that detects and mitigates advanced threats without compromising user privacy.

Main Methods:

• Utilizes adversarial federated learning with homomorphic inference for collaborative model improvement on encrypted traffic.
• Employs topological data analysis and Spiking Neural Networks (SNNs) on neuromorphic hardware for real-time anomaly detection.
• Integrates AI-assisted zero-knowledge attestations (ZK-SNARKs/STARKs) for verifiable client and resolver integrity.

Main Results:

• Achieved ≥99.20% accuracy in detecting malicious DNS activity.
• Demonstrated low latency for Zero-Knowledge (ZK) verification (<100ms) and handshake (<150ms).
• Federated learning models converged within 10 epochs, showcasing efficient training.

Conclusions:

• PAFA-DoH provides a practical, deployable solution for zero-trust, post-quantum hardened DNS protection.
• The combination of privacy-preserving AI, verifiable trust, and efficient analytics addresses critical security gaps in encrypted DNS.