Scoring Azure permissions with metric spaces

Christophe Parisel · 2025

Read Scoring Azure permissions with metric spaces by Christophe Parisel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-13747.txt.

Introduction

Parisel presents a paper from twenty twenty-five on scoring privilege risk in Microsoft Azure cloud accounts. Two scores cover the control side and the data side of the cloud. The control score orders accounts by their power to change settings. It weights write, action, and read rights in rising order. The data score gauges the widest reach of data theft or forgery. It leans on the tree shape of Azure tenant grouping. Together the scores form one frame for permission analysis.

Catalog record

CREATORChristophe Parisel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-13747

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Scoring Azure permissions with metric spaces Christophe Parisel Abstract In this work, we introduce two complementary metrics for quantifying and scoring privilege risk in Microsoft Azure. In the Control Plane, we define the WAR distance , a superincreasing distance over Write, Action, and Read control permissions, which yields a total ordering of principals by their configuration power. In the Data Plane, we present a blast radius distance for measuring the maximum breadth of data exfiltration and forgery, leveraging the natural ultrametry of Azure Tenants clustering hierarchy Together, these metrics offer a unified framework for proactive IAM analysis, ranking, lifecycle monitoring, and least privilege enforcement. 1 Introduction In modern cloud environments like Microsoft Azure, overseeing the security of privileged access is a complex and critical challenge, especially when it comes to Non Human Identities represented by Application SPNs and Managed Identities. As organizations move to the cloud, the need for robust systems to measure and mitigate risk associated with identity and access management (IAM) has become more pressing. One of the most significant risks is the potential for excessive or misconfigured privileges granted to NHIs aside of the standard groups membership process that human principals typically follow, which can lead to unauthorized data exfiltration, integrity breaches, or unintended privilege escalation. Azure provides fine-grained access control through role-based access control (RBAC) [ 1 ] , allowing principals (users, service principals, or managed identities) to be assigned permissions on resources across various scopes such as tenants, management groups (MG), subscriptions, resource groups (RG), and individual resources. However, understanding the exact risk posed by these permissions, particularly when they are distributed across different scopes and types, is a non-trivial problem. This paper attempts to solve that problem leveraging natural metric spaces found in Azure. We begin with an overview of the WAR distance metric, explaining how it combines scope and permission type to measure control-plane privilege. We then present the data-plane ultrametric for assessing risks related to data exfiltration and forgery. 2 Measuring Control Plane Privilege: The WAR Distance 2.1 Motivation Control plane permissions in Azure are assigned through role definitions containing actions . These define what operations a principal can perform on resources, including reading configurations, starting virtual machines, or deleting services. Each action is granted over a scope , such as a subscription or a resource group. To measure the privilege strength of any given Azure principal from a control-plane perspective, we introduce a distance function that reflects both the scope of permissions and their type . However, the space formed by combining scope and type is not totally ordered. The notion of a distance allows us to circumvent this limitation by imposing an opinionated but consistent ordering. 2.2 WAR: A Scope-Type Distance We define the WAR distance —short for Write/Action/Read—as a function that quantifies the distance between two sets of control-plane permissions. In particular, we are interested in computing the distance between a principal and the null origin , i.e., a principal with no assigned permissions. The WAR distance lives in the Cartesian product of two independent spaces: • Scope space : Ranging from tenant-level to sub-resource level: \texttt{Tenant}>\texttt{MG}>\texttt{Subscription}>\texttt{RG}>\texttt{Resource}>\texttt{Sub-resource} • Permission type space : We assume a control-plane risk-based order: \texttt{Write}>\texttt{Action}>\texttt{Read} 2.3 The Distance Model To define a distance that incorporates both scope and type, we assign numerical weights. Let each role assignment be modeled as a tuple (w,a,r)_{s} , where w , a , and r are integers representing the existence of write, action, and read permissions assigned at scope s ranging from Tenant to Sub-resource. The following discrete sets are used for the numerical weights: \displaystyle W \displaystyle=\{600,500,400,300,200,100\} \displaystyle A \displaystyle=\{90,80,70,60,40,20\} \displaystyle R \displaystyle=\{9,8,7,6,4,2\} The sets are chosen such that the minimum difference between any two elements in W (which is 100) is strictly greater than the maximum value in A plus the maximum in R (i.e., 90+9=99 ). Similarly, the smallest gap in A is 10, which exceeds the maximum value of R (which is 9). This superincreasing [ 4 ] structure guarantees that the total score can be decoded unambiguously into its components. Superadmin rights (denoted by wildcards in Azure permissions) are routinely assigned to buit-in roles like Owner or Contributor, and also to custom roles. At subscription level and above, this entails much more powerful rights than the Write permission. Taking wildcards into account, the permissions order becomes: \texttt{Wildcard}=>\texttt{Write}>\texttt{Action}>\texttt{Read} We also add three more weights to the Write set: one for Tenant superadmin (900), one for Management Group superadmin (800), and one for Subscription superadmin (700): \displaystyle W \displaystyle=\{900,800,700,600,500,400,300,200,100\} \displaystyle A \displaystyle=\{90,80,70,60,40,20\} \displaystyle R \displaystyle=\{9,8,7,6,4,2\} Whenever a wildcard is found at Subscription level or above, we replace the usual Write weight by its superadmin equivalent. 2.4 Privilege Norm and Principal Ranking We define the WAR distance between two points P_{1}=(w_{1},a_{1},r_{1}) and P_{2}=(w_{2},a_{2},r_{2}) as: d(P_{1},P_{2})=|w_{1}-w_{2}-a_{1}-a_{2}-r_{1}-r_{2}| This function, which we call the L1-minus distance , differs from the Manhattan (L1) distance. While the Manhattan distance is always a valid metric, the L1-minus is not point-separating in general. However, due to the superincreasing property of the sets W , A , and R , our version is indeed point-separating within our domain, and hence qualifies as a proper metric. Given a principal P=(w,a,r) , its privilege level (or strength) is: \|P\|=w+a+r Norm \|P\| induces by the WAR distance provides a scalar value to rank principals by privilege. The maximal norm is 999=900+90+9 , corresponding to a “superadmin” principal with full permissions at the tenant level. The null principal has \|P\|=0 . 2.5 Groups membership So far, we have only addressed the case of roles being granted directly to a principal. Principals often belong to groups, so they enjoy roles inheritance. To measure the actual norm of a principal belong to groups, we start by measuring its raw WAR norm, as we have just done, and we measure the norm of all groups (including nested ones). Consider the 3D vector of a WAR norm as a triangle with vertices W, A and R, centered at origin. The actual norm of the principal is the convex enveloppe of all triangles of the groups the principal belongs to, plus the triangle of its raw norm. For a concrete implementation, see [ 2 ] 2.6 Benefits of the WAR Distance The WAR distance is not only a tool for theoretical privilege comparison—it offers immediate operational benefits: 1. De-Escalation Planning: Principals can be sorted in decreasing order of \|P\| to identify the most powerful and risky SPNs and prepare targeted de-escalation strategies. 2. Target-Based Comparison: Given a principal’s current tuple and a target (less privileged) configuration, the WAR distance quantifies the privilege gap. This supports quantitative privilege reduction campaigns. 3. Lifecycle Monitoring and Anomaly Detection: By tracking \|P(t)\| over time, one can monitor the privilege evolution of any principal. Spikes or regressions in privilege level can be used to trigger alerts or to detect misconfigurations and privilege escalations. The WAR Distance is implemented in Azure Silhouette [ 2 ] , an open source tool developed by the author. 3 Measuring Data Plane Privilege: The Blast Radius Distance 3.1 Introduction In the context of data plane security, two fundamental risks dominate the threat landscape: data exfiltration and data forgery . These risks arise from overprivileged

Read the complete text