Researchers have identified a significant security vulnerability in Google's Vertex AI SDK that could have enabled attackers to hijack machine learning model uploads and execute code within Google's cloud infrastructure. The discovery underscores the evolving threat landscape in artificial intelligence development platforms where seemingly minor implementation flaws can lead to serious security consequences.
The vulnerability, discovered by Palo Alto Networks Unit 42 and reported through Google's bug bounty program, centered around an insecure implementation in the Python SDK for Google Cloud's Vertex AI service. Attackers could exploit a technique researchers termed "Pickle in the Middle" to intercept and compromise model uploads without requiring direct access to the victim's Google Cloud project. The attack worked through a method known as bucket squatting, where malicious actors could claim unused or misconfigured cloud storage buckets that the SDK would inadvertently use during the model upload process. Once in control of these storage