Top AI 2026 | MITRE ATLAS Mapped AML.T0043 – Adversarial Backdoor Trigger Injection Detection (AML-DET-007)
Detects attempts to trigger a poisoned machine learning model through anomalous inference requests. The rule identifies inference API calls flagged as statistically anomalous by an upstream detector (e.g., unusual pixel patterns, token sequences, or steganographic payloads) that are correlated with a specific, spike-related output class across multiple distinct users.
YARA-L
