The Agentic AI Framework (AAIF): a policy-enforced architecture for accountable and high-performance intrusion...

Area of Science:

• Cybersecurity
• Artificial Intelligence
• AI Governance

Background:

• Modern cybersecurity relies on AI for threat detection, but current systems often lack accountability, transparency, and policy compliance.
• Existing AI models prioritize detection accuracy, creating a gap in auditable and ethically aligned operations.

Purpose of the Study:

• To propose the Agentic AI Framework (AAIF), a novel policy-aware intrusion detection architecture.
• To integrate predictive AI modeling with executable governance for enhanced cybersecurity operations.
• To ensure AI decisions are auditable, transparent, and compliant with established frameworks like NIST AI RMF 2.0.

Main Methods:

• Design Science Research methodology.
• Integration of a deep learning detection model with a governance layer.
• Development of an interpretable Policy Engine using a YAML-based domain-specific language for constraint enforcement.

Main Results:

• AAIF maintained competitive predictive performance against baseline models (RF, SVM, DNN) with a weighted F1-score of 0.483 and AUROC of 0.978.
• Achieved complete policy compliance, including an Ethical Compliance Rate of 1.0 and a False Escalation Rate of 0.0.
• Improved the Governance Compliance Index from 0.947 to 0.983, indicating enhanced alignment with governance requirements.

Conclusions:

• Policy-enforced AI inference supports accountable autonomy in cybersecurity without compromising detection capabilities.
• The AAIF offers a reproducible, governance-aware approach, transforming intrusion detection systems into transparent and auditable decision-making tools.
• This framework provides a practical foundation for deploying policy-aligned AI in cybersecurity environments.