MITRE ATLAS 2026 Top AI Detection: Training Data Poisoning via Anomalous Dataset

Ibrahim Saud@tektrix

•updated 6 days ago•0•0•0

Detects anomalous data mutations in machine learning training pipelines, such as bulk insertions or label distribution shifts, performed by low-reputation or untrusted contributors. This activity is indicative of attempts to poison model training data to induce backdoors or skew decision boundaries.