MITRE ATLAS Mapped 2026 Top AI Adversarial Input Detection – Craft Adversarial Data (AML.T0043)

Ibrahim Saud@tektrix

•updated 23 days ago•0•0•1

Detects anomalous high-volume API requests to machine learning model inference endpoints (e.g., predict, classify, score). This behavior is characteristic of adversarial example crafting, where an attacker sends rapid, near-duplicate requests to estimate gradients or probe confidence scores to manipulate model output.