自動化數據管線上游污染

Risk Description

Attackers contaminate upstream sources of automated data collection pipelines (public datasets, RSS feeds, API endpoints) to inject harmful content at scale. As modern AI training heavily relies on automated pipelines, contaminated data can enter training sets without manual review. The impact far exceeds single dataset poisoning, potentially affecting models across multiple organizations simultaneously.