AI Artifact Collection
AI artifact collection is when a hacker steals the 'ingredients' used to build or run an AI, such as the AI's brain (the model), the information it learned from (the dataset), or logs of how it works, to either steal the technology or prepare for a future attack.
AI artifact collection involves the unauthorized acquisition of proprietary AI assets, including model weights, training datasets, and operational telemetry. Adversaries gather these components to facilitate exfiltration of intellectual property or to conduct reconnaissance and staging for subsequent adversarial machine learning attacks.
AI artifact collection is the tactical acquisition of machine learning-specific assets—encompassing model parameters, architecture, training/fine-tuning datasets, and inference-time telemetry—by an adversary. This activity serves as a prerequisite for exfiltration (AML.TA0010) or as a foundational step in AI attack staging (AML.TA0001), enabling the adversary to analyze model vulnerabilities, perform model inversion, or facilitate membership inference attacks.