Harden ML dependencies — hash-pinned lockfiles, allowlists, slopsquatting/typosquatting defense for PyTorch/TensorFlow/Transformers, private index prioritization against dependency confusion, Snyk/Grype/Trivy SCA, CycloneDX/SPDX SBOMs. Use when installing, auditing, or mirroring ML packages.

AI Dependency and Package Security

Implement ML model integrity using OpenSSF Model Signing with Sigstore transparency logs, SafeTensors/ONNX enforcement, pickle quarantine, SLSA provenance, immutable registries, ModelScan pre-ingest, trigger-input backdoor testing. Use when signing, verifying, or ingesting models.

AI Model Integrity and Provenance Security

Design end-to-end AI/ML supply chain security — SafeTensors over pickle, ModelScan plus OpenSSF Model Signing via Sigstore, slopsquatting/typosquatting defense, CycloneDX ML-BOM, cosign container signing, patched GPU runtimes. Use when securing model acquisition, deps, or deployment.

AI Supply Chain Security Architect

Harden training, fine-tuning, and RAG data against poisoning — authenticated sources with SHA-256 pinning, distribution anomaly and injection scanning, in-toto/SLSA attestations, backdoor trigger testing, RAG ACLs with rollback, isolated eval sets. Use when curating AI training or RAG corpora.