What problem does it solve?
It helps you identify security risks across the full lifecycle of AI/ML models—provenance, training data lineage, fine-tuning integrity, inference dependencies, documentation gaps, and backdoor signals—so you can produce verifiable, framework-mapped findings.
Core Features & Use Cases
- Model provenance verification: Checks whether model artifacts are sourced, pinned, and integrity-verified (e.g., checksums/signatures/attestations) rather than implicitly trusted from third-party registries.
- Training data lineage and governance review: Assesses poisoning exposure, data versioning, validation/deduplication, and access controls for datasets used in training/fine-tuning.
- Fine-tuning pipeline integrity: Evaluates SLSA-aligned build/training practices like reproducibility, provenance, credential separation, and signing/attestation of produced weights.
- Inference dependency and runtime risk review: Reviews serving-path dependencies and flags dangerous deserialization or insecure runtime patterns in the model deployment stack.
- Model card evaluation and backdoor detection patterns: Verifies whether documentation supports risk decisions and checks whether behavioral testing/monitoring covers backdoor threats.
Quick Start
Provide the target codebase path or artifact directory by telling the AI to review the model supply chain in the folder you specify.
