Google Cloud Vertex AI Cross-Tenant Remote Code Execution Vulnerability in Vertex AI Experiments

A vulnerability in Google Cloud Vertex AI Experiments, affecting versions 1.21.0 prior to 1.133.0, allows unauthenticated remote attackers to execute code across different tenants, steal models, and poison them. This is achieved by pre-creating predictably named Cloud Storage buckets, a tactic known as bucket squatting. The vulnerability has been patched, and no action is required from customers.

Impact

Exploitation of this vulnerability could lead to unauthorized cross-tenant remote code execution, model theft, and poisoning in Google Cloud Vertex AI.