Synopsis
Activation Period |
14 Days |
Training Period |
30 Days |
Test Period |
N/A (single event) |
Deduplication Period |
5 Days |
Required Data |
|
Detection Modules |
AIDR |
Detector Tags |
Cloud AI Infrastructure Analytics |
ATT&CK Tactic |
|
ATT&CK Technique |
|
Severity |
Low |
Description
AI knowledge base was modified by an identity that typically doesn't interact with knowledge bases.
MITRE ATLAS Technique: AML.T0070 - RAG Poisoning.
OWASP Top 10 LLM Technique: LLM04 - Data and Model Poisoning.
Attacker's Goals
Contaminating knowledge base, so that contextual information will be incorrect, biased or harmful.
Investigative actions
- Check the identity that modified the knowledge base.
- Check recent additions to the knowledge base.