https://taxonomy.eticas.ai/risk/data-poisoning
Maturity: established
Adversarial manipulation of training or fine-tuning data to introduce backdoors, bias, or systematic errors into the model. Includes both broad-spectrum poisoning (degrading overall performance) and targeted poisoning (creating exploitable triggers or specific failure modes).
Also known as: Training data poisoning · Backdoor attacks
System type: ADM and LLM systems
Lifecycle stages: Pre Processing, In Processing
| Framework | Reference |
|---|---|
| EU AI Act (Regulation 2024/1689) | Recital 76 — data poisoning (cybersecurity) |
| AIUC-1 — AI Underwriting Company Standard | Third-party testing of adversarial robustness |
| NIST AI 600-1 — Generative AI Risk Profile | Information Security (data poisoning) |
| NIST AI Risk Management Framework (AI 100-1) | Secure & Resilient (training-time attacks) |
| Framework | Reference |
|---|---|
| W3C Data Privacy Vocabulary — AI Extension | Data Poisoning |
| IBM AI Risk Atlas | Input → Data poisoning / Training-data tampering |
| MIT AI Risk Repository | AI system security vulnerabilities and attacks |