ADARA: Adversarial Deception-Aware Risk Architecture

ADARA (Adversarial Deception-Aware Risk Architecture) is a research prototype for detecting adversarial manipulation in AI-assisted command and control systems.

The system estimates the probability of adversarial deception across heterogeneous sensor inputs using a four-stage Bayesian cascade with Monte Carlo confidence intervals. The resulting deception estimate dynamically constrains autonomous authority through the relation:

A_adj = A_hmaa × (1 − λ × P_deception)

The software includes:

• Browser-executable prototype
• Canonical source code
• Build pipeline and CI verification gates
• 394 automated self-tests
• SHA-256 tamper-evident audit chain
• AES-256-GCM encrypted session export
• SIEM audit streaming
• Multi-operator RBAC enforcement

All sensor data used by the system is synthetic. The system must not be used in operational command and control contexts.

This deposit contains the complete reproducible artifact including source code, build scripts, test suite, and technical documentation.