An attacker can pre‑create predictably named Cloud Storage buckets to exploit predictable bucket naming in Google Cloud Vertex AI Experiments versions 1.21.0 to 1.133.0. By doing so, the attacker can gain cross‑tenant remote code execution, steal trained models, or poison models running on other tenants' data. The weakness is a predictable naming scheme that lacks proper authorization checks, documented as CWE‑340.

Google Cloud Vertex AI Experiments versions 1.21.0 through 1.133.0 (excluding 1.133.0) are affected. Any tenant using these versions is susceptible.

The CVSS base score is 7.7, indicating moderate severity, while the EPSS score is less than 1%, implying a very low likelihood of exploitation at the time of this analysis. The vulnerability is not listed in the CISA KEV catalog. Attacks can be launched without prior authentication by creating buckets with predictable names. Google has patched the issue; no customer action is required at this time.