Identity Risk Scoring

v5 status

In v5, risk scoring is driven from the UI (Admin > Risk Scoring). The PowerShell risk scoring functions (Invoke-FGRiskScoring, New-FGRiskProfile, New-FGRiskClassifiers) in tools/riskscoring/ are stubbed and return "not yet implemented in v5". The in-browser wizard and the Node.js scoring engine (app/api/src/riskscoring/engine.js) are the active implementation path.

Overview

Identity Atlas includes a universal risk scoring engine that assigns risk scores (0–100) to all entity types — Principals, Resources (including BusinessRoles), Contexts, and Identities. All scoring runs on your own infrastructure — no sensitive identity data is ever sent to external services.

Three-Phase Architecture

flowchart LR
    subgraph Phase1["Phase 1: Context (one-time)"]
        A["New-FGRiskProfile\nPublic domain info only"] --> B["New-FGRiskClassifiers\nIndustry-specific patterns"]
    end
    subgraph Phase2["Phase 2: Scoring (daily)"]
        B --> C["Invoke-FGRiskScoring\n4-layer batch engine"]
    end
    subgraph Phase3["Phase 3: Analysis"]
        C --> D[Risk Scoring UI Page]
        C --> E["Org Chart with\nrisk propagation"]
        C --> F[SQL: RiskScores table]
    end

Phase 1 (one-time setup):

New-FGRiskProfile -Domain "yourcompany.com" -LLMProvider Anthropic -LLMApiKey $key -ConfigFile '.\Config\mycompany.json'
New-FGRiskClassifiers -ConfigFile '.\Config\mycompany.json'

Phase 2 (run after each sync):

Invoke-FGRiskScoring -ConfigFile '.\Config\mycompany.json'

Phase 3 (analysis):

• UI: Risk Scoring page, Org Chart
• SQL: SELECT entityId, entityType, riskScore, riskTier FROM RiskScores ORDER BY riskScore DESC
• Fast filter: riskScore and riskTier columns are denormalized onto Principals and Resources

4-Layer Scoring Engine

flowchart TD
    E[Entity] --> L1["Layer 1: Direct Match\nClassifier patterns on name/description\nResource type multiplier"]
    L1 --> L2["Layer 2: Membership Analysis\nHigh-risk member/group density\nPIM eligibility, no-owner signals"]
    L2 --> L3["Layer 3: Structural Hygiene\nStale sign-in, never used\nGuest accounts, missing description\nDynamic membership, directory role signals"]
    L3 --> L4["Layer 4: Propagation\nInherit risk from riskiest\nmember / group / parent"]
    L4 --> S["Final Score 0–100\nClamped + tiered"]

Each layer adds points; the final score is clamped to 0–100.

Scoring by Entity Type

Entity Type	Layer 1	Layer 2	Layer 3	Layer 4
Principal (User)	Classifier matches on name/title/UPN	Group count, PIM eligibility, high-risk group density	Stale sign-in (90/180d), never signed in, guest account, active high-privilege usage	Riskiest group membership
Principal (AI Agent/SP)	Agent classifier matches on name	Group count, high-risk resource density	No-human-in-loop penalty, type-specific bonus, active production workload detection	Riskiest resource membership
Resource	Classifier + resourceType multiplier	High-risk member density, PIM-eligible members, no owner	No description, dynamic membership, directory role signals, ghost app roles	Riskiest member
BusinessRole (Resource)	Classifier matches	High-risk assignee density	No review configured, missed reviews, no auto-removal	Aggregate contained resource risk
OrgUnit	Name/dept classifiers, hierarchy position	Aggregate principal risk	Size extremes (tiny/huge), no manager, high external ratio	Parent OrgUnit risk
Identity	Account count, multi-system	Highest-risk linked principal	Orphaned accounts, low confidence, not verified	Critical/High principal propagation

Risk Tiers

Tier	Score	Action
Critical	90–100	Requires immediate attention
High	70–89	Should be reviewed soon
Medium	40–69	Monitor regularly
Low	20–39	Low concern
Minimal	1–19	Negligible
None	0	No signals detected

Analyst Overrides

Analysts can adjust any entity's score with a required justification. The adjustment must be a whole number between −50 and +50, and the justification must be 3–500 characters:

• Via UI: Risk Scoring page → click Override button
• Overrides are stored in RiskScores and preserved across re-scoring runs
• Override history is preserved across re-scoring runs

Configuring the AI provider

The AI provider is configured in Admin → LLM Settings. Three providers are supported:

Provider	provider value	Default model	Notes
Anthropic	anthropic	claude-sonnet-4-6	API key required
OpenAI	openai	gpt-4o	API key required
Azure OpenAI	azure-openai	(deployment name required)	Also requires endpoint URL and deployment name

Once you enter the API key and click Test, IdentityAtlas verifies the connection and populates the model dropdown from the provider's API. The key is stored encrypted in the built-in secrets vault — it is never stored as plain text.

For Azure OpenAI the endpoint must be a valid Azure OpenAI hostname (*.openai.azure.com, .us, or .cn). The deployment name is the name you gave the deployment in Azure, not the underlying model name.

Data Privacy

No identity data ever leaves your infrastructure

• Phase 1 only contacts an LLM — the prompt contains only public organisational context: the domain name, optional organisation name, optional free-text hints you type, and any text you paste from internal documents.
• No identity data — no user names, email addresses, group names, or access assignments — is ever sent to the AI provider.
• All scoring (matching classifiers against your data) runs locally against your database with no network calls.

What is sent to the AI, exactly:

What	Example
Domain name	contoso.com
Organisation name (optional)	Contoso Ltd
Free-text hints (optional)	We operate in the Netherlands under NIS2
Text from URLs you add (optional)	Scraped text from public or internal pages

Nothing else. The AI generates regex patterns from this context; those patterns are then stored locally and applied against your data entirely on-premises.

Provider security comparison:

Provider	Data stays in your region?	Used for AI training?
Anthropic API	No — routed to api.anthropic.com	Depends on your plan. Paid API plans include a no-training clause; verify your agreement.
OpenAI API	No — routed to api.openai.com	Consumer tier: opt-out required in account settings. Enterprise agreement: excluded by contract.
Azure OpenAI	Yes — stays within your Azure subscription	No — Microsoft Azure terms contractually exclude customer prompts from model training.

For regulated environments or when in doubt, use Azure OpenAI: the data stays within your Azure tenant and training exclusion is contractual by default.

RiskScores Table

-- Top risky principals
SELECT p.displayName, p.principalType, rs.riskScore, rs.riskTier,
       rs.directScore, rs.membershipScore, rs.structuralScore, rs.propagatedScore
FROM Principals p
JOIN RiskScores rs ON rs.entityId = p.id AND rs.entityType = 'Principal'
ORDER BY rs.riskScore DESC;

-- All entity types together
SELECT entityId, entityType, riskScore, riskTier
FROM RiskScores
ORDER BY riskScore DESC;