Back to controls

AI models should not be reachable through broad identities

AI models and endpoints should run with least privilege. Broad identities can allow model extraction, unauthorized deployment changes, data access through tools, or lateral movement into storage and vector databases used by AI workloads.

Category

Controls

High

Applies to

AWSGoogle CloudKubernetesMicrosoft Azure

Coverage

1 queries

Asset types

3 covered

Overview

AI models and endpoints should run with least privilege. Broad identities can allow model extraction, unauthorized deployment changes, data access through tools, or lateral movement into storage and vector databases used by AI workloads.

Remediation guidance

Remediation

Replace broad model and endpoint identities with workload-specific roles. Grant only the actions required to invoke, deploy, or read the specific models and datasets used by the workload.

Rollout guidance

  1. Inventory the identities used by AI endpoints and training jobs.
  2. Replace owner, contributor, editor, administrator, and wildcard policies with scoped permissions.
  3. Separate deployment identities from runtime inference identities.
  4. Re-scan and track temporary exceptions with expiry dates.

Query logic

These are the stored checks tied to this control.

AI models should not be reachable through broad identities

Connectors

AWSGoogle CloudKubernetesMicrosoft Azure

Covered asset types

AI ServicesIAM RoleService Account

Expected check: eq []

{
  sageMakerModels(where: { executionRoleARN_MATCHES: ".*:role/(admin|administrator|owner|power|full).*" }) { ...AssetFragment }
  vertexAIEndpoints(where: { deployedModels_SOME: { serviceAccount_MATCHES: "(?i).*(admin|owner|editor|power).*" } }) { ...AssetFragment }
  azureAIServiceAccounts(where: { OR: [ { roleIDs_INCLUDES: "Owner" } { roleIDs_INCLUDES: "Contributor" } { roleIDs_INCLUDES: "User Access Administrator" } { roleIDs_INCLUDES: "Administrator" } ] }) { ...AssetFragment }
  azureAIFoundryProjects(where: { OR: [ { roleIDs_INCLUDES: "Owner" } { roleIDs_INCLUDES: "Contributor" } { roleIDs_INCLUDES: "User Access Administrator" } { roleIDs_INCLUDES: "Administrator" } ] }) { ...AssetFragment }
  azureOpenAIDeployments(where: { OR: [ { roleIDs_INCLUDES: "Owner" } { roleIDs_INCLUDES: "Contributor" } { roleIDs_INCLUDES: "User Access Administrator" } { roleIDs_INCLUDES: "Administrator" } ] }) { ...AssetFragment }
  azureMachineLearningEndpoints(where: { OR: [ { roleIDs_INCLUDES: "Owner" } { roleIDs_INCLUDES: "Contributor" } { roleIDs_INCLUDES: "User Access Administrator" } { roleIDs_INCLUDES: "Administrator" } ] }) { ...AssetFragment }
}
Cyscale Logo
Cyscale is an agentless cloud-native application protection platform (CNAPP) that automates the contextual analysis of cloud misconfigurations, vulnerabilities, access, and data, to provide an accurate and actionable assessment of risk.

Stay connected

Receive new blog posts and product updates from Cyscale

By clicking Subscribe, I agree to Cyscale’s Privacy Policy


© 2026 Cyscale Limited

LinkedIn icon
Twitter icon
Facebook icon
crunch base icon
angel icon