Definition
An attacker preemptively creates a cloud storage bucket with a predictable name (often derived from a victim's project ID or region) before the legitimate user attempts to upload data to it. When the legitimate process attempts the upload, files go into the attacker's bucket instead, enabling model poisoning, credential theft, or code injection.
Why it matters
In machine learning pipelines, bucket squatting enables cross-tenant compromise without requiring access to the victim's account. An attacker can poison a model before it reaches production, causing it to behave maliciously when deployed—silently and without any sign of compromise.