Entity Risk Assessment

Agent, system, and fleet-level risk scoring across all behavioral history

Superseded by Intent-Scoped Security (2026-07-18). Per-agent / per-fleet behavioral baselining (EWMA of daily scores, action entropy, trend trajectory) is the pre-pivot substrate. It is now shadow-only scaffolding that makes no enforcement decision and is retired only on Phase 3 A/B evidence; detection is driven by deterministic scope enforcement and intent-vs-truth divergence, not by an entity risk score.

Entity Risk Assessment

Beyond per-event scoring, Quint evaluates the overall risk posture of agents, systems, and entire fleets. Entity assessment uses 8 dimensions computed from the complete behavioral graph.

Assessment Levels

Level 4: Fleet Assessment     "How risky is our entire agent infrastructure?"
Level 3: System Assessment    "How risky is this application/platform?"
Level 2: Agent Assessment     "How risky is this specific agent?"
Level 1: Session Assessment   "How risky was this particular session?"

Agent Assessment (Level 2)

The most common assessment. Produces scores across 8 dimensions, each 0-100.

Dimensions

Dimension	Weight	Description
Historical Event Risk	0.20	EWMA of daily max scores (recent events weight more)
Privilege Footprint	0.15	Blast radius: resources, fields, servers, OAuth scopes
Behavioral Consistency	0.15	Action entropy, session variance, resource stability
Policy Compliance Rate	0.15	Violation frequency weighted by severity
Data Sensitivity Exposure	0.10	Highest-sensitivity data regularly accessed
External Communication Risk	0.10	Outbound data volume, unwhitelisted endpoints
MCP Trust Surface	0.08	Unverified servers, description changes, broad schemas
Trend Trajectory	0.07	Is the agent getting riskier or safer?

Agent Composite Formula

agent_risk = (
  0.20 × historical_event_risk +
  0.15 × privilege_footprint +
  0.15 × behavioral_consistency +
  0.15 × policy_compliance +
  0.10 × data_sensitivity_exposure +
  0.10 × external_communication_risk +
  0.08 × mcp_trust_surface +
  0.07 × trend_trajectory
)

Example Agent Profile

{
  "agent_id": "support_bot_v3",
  "agent_risk_score": 34,
  "risk_level": "medium",
  "dimensions": {
    "historical_event_risk": {"score": 22, "weight": 0.20},
    "privilege_footprint": {"score": 58, "weight": 0.15},
    "behavioral_consistency": {"score": 15, "weight": 0.15},
    "policy_compliance": {"score": 8, "weight": 0.15},
    "data_sensitivity_exposure": {"score": 45, "weight": 0.10},
    "external_communication_risk": {"score": 62, "weight": 0.10},
    "mcp_trust_surface": {"score": 10, "weight": 0.08},
    "trend_trajectory": {"score": 38, "weight": 0.07}
  },
  "risk_drivers": [
    {
      "dimension": "external_communication_risk",
      "score": 62,
      "explanation": "Sends data to 3 external endpoints including 1 not on whitelist",
      "recommendation": "Review and whitelist api.analytics-partner.io"
    }
  ],
  "trend": {
    "7d_avg": 18,
    "14d_avg": 21,
    "30d_avg": 22,
    "direction": "improving"
  }
}

System Assessment (Level 3)

Aggregates agent scores and adds cross-agent signals:

Signal	Weight	Description
Weighted agent scores	0.50	Weighted by action volume
Cross-agent resource contention	0.20	Multiple agents accessing same sensitive resources
Shared MCP concentration	0.15	Single-point-of-compromise MCP servers
Delegation chain risk	0.15	Cross-agent delegation depth

Fleet Assessment (Level 4)

CISO dashboard view across all agents and systems:

{
  "fleet_risk_score": 31,
  "total_agents": 23,
  "total_systems": 4,
  "systems": [
    {"system": "customer_support", "score": 38, "agents": 7},
    {"system": "code_review", "score": 22, "agents": 5},
    {"system": "data_pipeline", "score": 45, "agents": 8},
    {"system": "sales_outreach", "score": 18, "agents": 3}
  ],
  "fleet_stats": {
    "total_actions_30d": 2847000,
    "total_policy_violations_30d": 342,
    "violation_rate": "0.012%"
  }
}

Entity Risk Levels

Score	Level	Meaning
1-15	Low	Operating within baseline
16-35	Moderate	Some drift, monitor
36-60	Elevated	Significant changes, investigate
61-80	High	Active risk, restrict permissions
81-100	Critical	Likely compromised, immediate action

Assessment Triggers

Trigger	Frequency
Scheduled (background)	Agent: 6h, System: 12h, Fleet: 24h
Threshold-triggered	On any event exceeding alert threshold
API-requested (on-demand)	GET /v1/assess/agent/{id}?window=30d