Deep anomaly detection for active attacks on the receiver in quantum key distribution

Area of Science:

• Quantum Information Science
• Cybersecurity
• Machine Learning

Background:

• Traditional defenses against quantum key distribution (QKD) receiver attacks face infrastructure compatibility issues and limited scope.
• Existing methods may introduce new vulnerabilities or fail to detect diverse active attack types.

Purpose of the Study:

• To develop and evaluate an anomaly detection (AD) model for safeguarding QKD receivers against active attacks.
• To provide a cost-effective and broadly applicable solution for enhancing QKD security.

Main Methods:

• Constructed a dataset from QKD system operational states to train a one-class machine learning AD model.
• The AD model learns normal system behavior and identifies deviations indicative of active attacks.

Main Results:

• The AD model achieved an area under the curve (AUC) exceeding 99%, demonstrating high detection accuracy.
• The model effectively safeguards QKD receivers by identifying anomalous operational states during attacks.

Conclusions:

• The proposed AD model offers a practical and accurate solution for active attack detection in QKD systems.
• It is easily deployable in existing infrastructure with minimal cost and avoids introducing new side channels.
• The model's ability to detect unknown attacks enhances its versatility beyond traditional, attack-specific methods.