Paper Type

Complete

Paper Number

PACIS2025-1787

Description

Quantum computing threatens classical encryption, exposing financial transactions to future “harvest-now, decrypt-later” attacks. We present a real-time security framework that fuses AI-driven anomaly detection with adaptive, post-quantum key management to protect high-volume financial streams. An Autoencoder monitors live transaction data, flagging irregular patterns with 97 % detection accuracy and < 50 ms latency. Detected anomalies trigger a policy engine that rotates or escalates cryptographic keys from RSA/ECC to CRYSTALS-Kyber/Dilithium within 35 ms, enabling cryptographic agility without disrupting service. Using an enriched FFIEC dataset of 150 000 synthetic transactions containing 5 % injected fraud, the combined system reduced false positives relative to signature-based IDS while sustaining throughput on commodity servers. Our results demonstrate that integrating machine-learning “AI-sensing” with quantum-safe cryptography provides a scalable defence against both current and imminent quantum-enabled threats, offering actionable design guidance for financial institutions preparing for the hybrid quantum-classical era of secure operations.

Comments

Blockchain

Share

COinS
 
Jul 6th, 12:00 AM

Post-Quantum AI-Driven Cryptographic Key Management for Financial Anomaly Detection

Quantum computing threatens classical encryption, exposing financial transactions to future “harvest-now, decrypt-later” attacks. We present a real-time security framework that fuses AI-driven anomaly detection with adaptive, post-quantum key management to protect high-volume financial streams. An Autoencoder monitors live transaction data, flagging irregular patterns with 97 % detection accuracy and < 50 ms latency. Detected anomalies trigger a policy engine that rotates or escalates cryptographic keys from RSA/ECC to CRYSTALS-Kyber/Dilithium within 35 ms, enabling cryptographic agility without disrupting service. Using an enriched FFIEC dataset of 150 000 synthetic transactions containing 5 % injected fraud, the combined system reduced false positives relative to signature-based IDS while sustaining throughput on commodity servers. Our results demonstrate that integrating machine-learning “AI-sensing” with quantum-safe cryptography provides a scalable defence against both current and imminent quantum-enabled threats, offering actionable design guidance for financial institutions preparing for the hybrid quantum-classical era of secure operations.