Unit 42 identified a critical vulnerability in the Vertex AI Python SDK where an attacker can perform bucket squatting to hijack model uploads, leading to cross-tenant remote code execution (RCE). This vulnerability exploits the handling of model uploads to cloud storage buckets, allowing an attacker to inject malicious code that executes remotely in the context of the victim tenant. The detailed technical analysis is available in the Unit 42 article. No explicit patch or remediation details are provided in the current data.

Successful exploitation of this vulnerability results in remote code execution across tenants in the Vertex AI environment, which can lead to unauthorized control over victim resources and compromise of data confidentiality and integrity. The attack leverages bucket squatting to intercept and manipulate model uploads, posing a critical security risk to affected users.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is released, users should monitor vendor communications and consider restricting permissions related to bucket creation and model uploads where possible. Avoid using untrusted or shared buckets for model uploads to reduce exposure.