Skip to content
The principal risk score

The principal risk score (UPR)

Unified Principal Risk (UPR) gives you a 0–100 score for each principal. See the contribution from eight posture factors, active privileged sign-in risk and open findings, then investigate the access behind the score.

One score for the whole principal

A principal can hold a broad Azure role, carry sensitive application permissions and have no recent sign-in. Each fact describes a different part of its exposure. UPR combines those signals so you can compare principals and choose where to investigate first.

The factor breakdown shows how privilege, access and protective controls contribute. A high score directs you to the underlying findings; a lower score reflects fewer or weaker risk contributions. Use the evidence behind the number to choose a change.

How UPR is computed

UPR starts with eight posture factors. Their signed contributions form the posture base, so protective controls reduce the total. The principal detail page shows the persisted breakdown.

  • Privilege level. Scores the highest privilege tier, from ordinary access to control of the directory or cloud. The contribution ranges from 0 to 30.
  • Permission creep. Maps the unused-permission input to a contribution from 5 to 25. Where effective-permission data exists, the input is the rounded percentage of assigned effective permissions without a matching observed operation: round(unused / assigned × 100).
  • Dormancy. Adds up to 20 for a principal without recent activity. This highlights access retained by inactive identities.
  • Scope breadth. Adds up to 15 according to the reach of the assignment. Broad scopes contribute more than narrow ones.
  • Sensitive permissions. Adds up to 10 for high-impact capabilities such as credential or role management.
  • PIM posture. Distinguishes standing privilege from eligible access. The contribution ranges from −10 to +5.
  • Conditional Access coverage. Applies protective credit or a penalty for user sign-in controls, from −15 to +10. This factor is neutral for non-human identities.
  • Disabled-account state. Subtracts 40 when the account is disabled. The reduction is applied before the final score is clamped.

Active Entra ID Protection sign-in risk adds 8 for medium risk or 12 for high risk on a privileged principal. The matching risky-sign-in finding is excluded from the findings contribution to avoid counting the same signal twice. The signal is recalculated on each analysis.

Open principal findings add 12 for each critical finding, 7 for high, 3 for medium and 1 for low, capped at 25 in total. UPR is the sum of the posture base, privileged sign-in risk and findings contribution, rounded and clamped to 0–100.

Distribution

Sample data

Lower UPR = less standing risk94 principals

Every factor traces to evidence

Open the principal detail page to inspect its factor breakdown. Privilege level comes from collected roles; scope breadth comes from assignment scopes. Dormancy uses activity and sign-in history. PIM and Conditional Access factors use the collected configuration.

Activity-based findings cite collected Azure activity-log evidence; configuration findings cite collected configuration. The distinction matters: a broad role assignment is a configuration fact, while observed use is a statement about activity inside the analysis window.

The unused-permission input compares effective Azure RBAC permissions with observed control-plane operations over 90 days. Use the role, scope and operation details together when preparing a least-privilege change. The Coverage section explains how to interpret this input.

How to read a UPR score

Principal risk levels use fixed thresholds. The level describes the combined score, not the severity of any single finding.

  • Critical: 80–100. The strongest concentration of risk factors. Inspect the largest contributions and open findings first.
  • High: 60–79. A substantial risk contribution from privilege, unused access, posture or findings. Plan the relevant access and configuration changes.
  • Medium: 40–59. Inspect the factor mix and track changes through your access-review cycle.
  • Low: 0–39. A lower combined contribution. You can still inspect individual findings and the controls reducing the score.

The tenant distribution uses a separate three-band scale: low 0–33, medium 34–67 and high 68–100. These distribution buckets are distinct from the four principal risk levels above. Both sets of thresholds are fixed.

What Permafrost does with high-UPR principals

  • Prioritization. Use identity scores and the privileged-set view to focus your investigation. Open a principal to see the factors and findings behind its score.
  • Right-sized role recommendations. For Azure RBAC permission gaps, inspect the recommended role and its observed operations. Export the role definition as ARM, Bicep or Terraform for your change process. Include required seasonal or recovery operations when assessing the proposed role.
  • Remediation. Choose among the modes supported by the finding and action: manual guidance, a downloadable script or a delegated in-product action. The full coverage reference is at /docs/three-mode-remediation.

Tenant aggregation

Tenant UPR is a blast-radius-weighted mean of principal scores plus a capped contribution from open tenant findings. Groups and removed records are excluded. Principal findings already contribute to each principal score and are not added a second time.

Each principal starts with weight 1. A privilege factor of 30 adds 4; otherwise the privileged flag adds 2. Scope breadth of 15 adds 2; scope breadth of 10–14 adds 1. A principal with the broadest scope and highest privilege therefore has weight 7.

Open tenant findings use the same severity weights as principal findings, with a cap of 15. The weighted mean plus that contribution is rounded and clamped to 0–100. The UPR API returns stored snapshots and a customer rollup; reading it does not trigger a fresh analysis.

Coverage

Azure activity logs record control-plane operations, not complete data-plane use. The initial collection requests 90 days of history; incremental collection resumes from the latest stored event with a two-hour overlap. Available history and granted read permissions determine the evidence collected.

When effective-permission rows are absent, the analyzer uses an identity-type heuristic for the unused-permission input. That input is not a measured percentage. A missing activity event does not establish that a workload has no need for the permission.

UPR is a prioritization score, not a probability of compromise or a percentage of access that can safely be removed. Configuration, activity and protective factors each describe a different part of the principal’s posture.

Start scoring your principals

Connect your Microsoft tenant with read-only access. After sync and analysis, open the identity inventory to inspect scores, contributing factors and findings. Use subsequent syncs to track changes.