A Unified Framework for Defending Against AI-Driven APTs
AI-enabled advanced persistent threats can automate parts of reconnaissance, evasion and attack adaptation. The paper proposes a defensive architecture that combines multiple controls instead of relying on any single detection method.
Threat intelligence
Blockchain-backed sharing is used to make collaborative threat intelligence more tamper-resistant.
Zero trust
Continuous verification of users, devices and access requests reduces implicit trust inside the environment.
Privacy-preserving analytics
Homomorphic-encryption concepts are incorporated to support analysis while reducing exposure of sensitive data.
Adversarial ML defense
The architecture includes defenses intended to improve resilience against machine-learning evasion and manipulation.
AI behavior auditing
Monitoring and auditing of AI-agent actions adds oversight for autonomous or semi-autonomous behavior.
Unified architecture
The central idea is that these controls reinforce one another when deployed as a cohesive defensive system.
Citation
Sim, Hyun and Inari, Vera. “A Unified Framework for Defending Against AI-Driven APTs.” April 11, 2025. SSRN. DOI: 10.2139/ssrn.5213825.