CyberCertPrepAI Security FundamentalsAdversarial ML & Model AttacksEASYFree question
A model training pipeline ingests data from 50 external sources via automated pipelines. A security engineer proposes implementing "data provenance tracking." What does this mean and why is it security-relevant for ML?
- A.A. Tracking when model files were last modified on disk
- B.D. Logging which GPU was used to train each batch
- C.C. Documenting the model's hyperparameters for reproducibility
- D.B. Recording the origin, transformation history, and custody chain of every training sample so that if a poisoning attack is detected, the affected data can be traced back to its source and quarantined
Why D is correct
Data provenance tracks the full lineage of each training sample: its source URL or database, when it was collected, which preprocessing functions transformed it, and which humans or automated systems touched it. For ML security, this enables: (1) rapid forensic investigation after a poisoning incident, (2) targeted quarantine of data from a compromised source without reprocessing the entire dataset, and (3) compliance documentation. Hyperparameters are experiment metadata. GPU logs and file modification times are operational metrics irrelevant to poisoning investigations.
Know someone studying for AI Security Fundamentals? Send them this one.