Risk Scoring Model
How ShieldAgent computes a 0–100 risk score per agent, what each tier means, and how enforcement decisions are made.
Overview
Every agent monitored by ShieldAgent has a continuous risk score between 0 and 100. The score is computed from security events, compliance gaps, integrity checks, and operational signals. Recent events contribute more to the score than older ones.
Tenant risk enforcement is off by default. Once enabled, actions depend on the effective tier after overrides and any grace period. Configured tenant rate limits and other proxy controls continue to apply at every tier.
Risk Tiers
| Tier | Score | Enforcement |
|---|---|---|
| Normal | 0 – 59 | No additional risk-tier restriction. Configured tenant rate limits and other proxy controls still apply. |
| Elevated | 60 – 79 | With risk enforcement enabled, an effective Elevated tier after overrides and grace reduces the configured tenant limit through the earlier rate-limit check. Pre-existing shadow mode observes over-limit checks without blocking. |
| High | 80 – 89 | With risk enforcement enabled, an effective High tier after overrides and grace reduces the tenant limit at the earlier rate-limit check. High switches the later risk stage to shadow; that does not undo an earlier enforced rate-limit denial. |
| Critical | 90 – 100 | With risk enforcement enabled, an effective Critical tier blocks non-allowlisted methods in enforce mode. Shadow mode records would-block decisions; allowlisted lifecycle methods remain subject to other controls. |
How Scoring Works
Risk scores combine signals across four categories: security events, compliance violations, integrity checks, and operational patterns. Each category is scored independently. Recent events contribute more than older ones, so a clean period naturally reduces the score without manual intervention.
What Drives the Score
Threat Signals
Detections like prompt injection attempts and data loss events raise the score. More severe threats have a greater impact.
Policy Violations
Repeated policy denials indicate an agent operating outside its intended boundaries.
Tool Integrity
Tool drift and supply-chain changes signal that the tools an agent relies on may have been tampered with.
Operational Patterns
Frequent human-in-the-loop triggers suggest an agent may be misconfigured or testing its boundaries.
Automatic Recovery
Risk scores naturally decrease over time as an agent operates cleanly. Recent events carry more weight than older ones, so a period of normal behavior brings the score back down without manual intervention.
Viewing Risk Scores in the Dashboard
The dashboard is the primary way to monitor and act on agent risk:
- 1Go to the Overview page — the top section shows agents sorted by risk tier with color-coded badges.
- 2Click any agent to open its detail page. The Risk tab shows the current score, tier, trend graph, and contributing events.
- 3For Critical-tier agents (score ≥ 90), a Release button appears — click it to restore normal operation for 24 hours while you investigate.
- 4Use the Agent Radar (topology view) with the Risk lens to see all agents color-coded by tier at a glance.