AEGIS Risk Scoring Model

Architectural Enforcement & Governance of Intelligent Systems

Version: 0.2
Status: Informational
Part of: AEGIS Architecture
Author: Kenneth Tannenbaum
Last Updated: March 6, 2026

Purpose

This model defines how risk is represented conceptually and mapped to governance outcomes. Numerical implementation details are specified in:

• docs/architecture/RISK_SCORING_ALGORITHM.md

Risk Model Dimensions

Risk is determined by five dimensions:

1. Actor trust posture.
2. Capability intrinsic risk.
3. Resource sensitivity.
4. Environment modifier.
5. Behavioral history modifier.

These dimensions provide contextual risk beyond static permissions.

Risk Bands

Conceptual Risk Factors

Actor Trust

• Reflects actor maturity, identity confidence, and prior behavior.

Capability Risk

• Reflects inherent potential impact of requested operation.

Resource Sensitivity

• Reflects classification and criticality of target resource.

Environment Modifier

• Reflects deployment context (production, off-hours, incident state).

History Modifier

• Reflects trend-based behavior and prior violations.

Risk-to-Governance Mapping¹

Risk is advisory to policy, but binding to outcome thresholds.

• Policies can lower or raise effective risk within bounded limits.
• Risk cannot override explicit deny policy.
• Critical risk cannot be silently downgraded to allow.

Model Invariants

1. Risk score must remain in bounded range [0, 100].
2. Same risk inputs must produce same score.
3. Missing high-impact factors cannot default to low risk.
4. High and critical bands must produce non-allow outcomes.

Operational Uses

Risk outputs are used for:

• Real-time decision classification.
• Escalation routing.
• Adaptive constraints (rate, timeout, scope).
• Trend analytics and post-incident tuning.

Calibration and Drift Control

Model maintenance requirements:

• Periodic calibration against incident outcomes.
• Drift detection for sudden score distribution shifts.
• Bias review across actor classes and environments.

Verification Criteria

The risk model is considered healthy when:

• Threshold transition tests pass at all boundaries.
• Risk distribution aligns with expected operational profile.
• High-risk events correlate with stricter governance outcomes.
• No critical-risk request is executed without escalation override.

References