Issue Details
Data and Model Poisoning
llm04-2025 • Rank #4 in OWASP 2025
Explanation
Data and model poisoning introduces malicious influence into training, fine-tuning, or retrieval corpora to bias outputs or implant hidden triggers. Attackers can shape responses without direct system compromise by contaminating trusted data paths. Detection is difficult because effects may appear gradual or context-specific.
Remedy
Implement provenance controls, integrity validation, and anomaly screening across all data ingestion and training workflows. Segment trusted and untrusted corpora and apply quarantine processes for suspicious samples. Continuously evaluate model behavior for drift and trigger patterns using adversarial testing.
| Year | Rank | Category |
|---|---|---|
| 2025 | #4 | llm04-2025 - Data and Model Poisoning |