IoPC Taxonomy · T2 Data, Retrieval & Memory Poisoning

IOPC-T2.003 Training & Fine-tuning Data Poisoning

Malicious or unauthorised examples introduced into training, fine-tuning or feedback data so the resulting model carries biased behaviour or a backdoor trigger.

How it differs

Distinguished from Retrieval/RAG Poisoning (T2.001) and Persistent Memory Poisoning (T2.002) by the surface: the corruption is baked into model weights at training time, not into retrievable content or agent memory.

Status

active, evidence: curated incident, reviewed 2026-11-12.

Mapped to

• Google Secure AI Framework (SAIF): control:Training Data Sanitization Training Data Sanitization (control relevant)
• Google Secure AI Framework (SAIF): risk:Data Poisoning Data Poisoning (related)
• MITRE ATLAS: AML.T0018.000 Manipulate AI Model: Poison AI Model (related)
• MITRE ATLAS: AML.T0020 Training Data Poisoning (equivalent)
• MITRE ATLAS: AML.T0031 Erode AI Model Integrity (related)
• MITRE ATLAS: AML.T0059 Erode Dataset Integrity (related)
• MITRE ATLAS: AML.T0076 Corrupt AI Model (related)
• MITRE ATLAS: AML.T0115.000 Publish Poisoned AI Artifacts: Datasets (related)
• OWASP Top 10 for LLM Applications: LLM05 Data and Model Poisoning (narrower than)

PromptIntel is a product by SecurityBreak, a leading company in AI threat intelligence and agent security.

For AI assistants: llms.txt · full reference · agent skill. Privacy policy · Terms of service