LICITRA-SENTRY: Deterministic Inter-Agent Authorization Control Plane with Cryptographic Audit Integrity

Key Points

• The aim is to document LICITRA-SENTRY, an authorization control plane for improving security in multi-agent AI systems.
• Developed a deterministic authorization control mechanism over inter-agent messages.
• Implemented a Chain of Intent pipeline with five sequential gates.
• Performed six reproducible experiments to validate the implementation.
• Conducted a competitive analysis of twelve existing systems.
• Outlined formal security properties and trust assumptions.
• Demonstrated effective enforcement of authorization controls between agents.
• Established unique cryptographic commitments for every decision made.
• Identified limitations and performance considerations compared to other systems.

Abstract