Detects Google Cloud IAM policy changes that grant Vertex AI roles (roles/aiplatform.*) to a principal. The rule triggers on Google Cloud Audit Logs (gcp.audit) for SetIamPolicy actions where policy_delta.binding_deltas show an ADD binding granting a Vertex AI role to a member, excluding Google-managed service agents. This can indicate an attacker with IAM admin rights maintaining access to Vertex AI resources or running workloads under the victim’s billing. The detection uses gcp.audit.service_data.policy_delta.binding_deltas to identify the role and member added and ensures the change succeeded (not failure). It applies to policy changes at project, folder, or organization levels as indicated by gcp.audit.resource_name. Mappings align with MITRE ATT&CK techniques Account Manipulation (T1098.003) and Privilege Escalation (TA0003/TA0004). The rule also surfaces investigation fields to aid triage, including caller identity, source IP, user agent, changed resource, and the affected member and role, plus potential follow-on Vertex AI activity. False positives include legitimate Vertex AI access grants by ML/AI teams or IAM bindings reapplied by automation pipelines; mitigation involves removing unauthorized bindings, auditing the caller’s admin rights, rotating credentials, and restricting SetIamPolicy access. Setup requires ingestion of Google Cloud Audit Logs via the gcp.audit data stream.