ml-security

The model and data artifacts: checkpoint formats that execute code on load, provenance for a model you did not train, training-data poisoning and the ingestion controls that bound it, PII that survives into weights, and notebooks that commit their own output. Use when loading a model from disk, a Hu