End-to-End Secure ML Development
summary
Covers an end-to-end recipe that actually signs and verifies models using model signing with six store and logs inclusion proofs so model hubs can verify offline; demonstrates a 1 TB model hash costing ~900 seconds and a demo hash taking ~17 seconds, and uses 10‑minute ephemeral certificates to limit key theft risk. Shows a GitHub Actions proof‑of‑concept that emits SLSA provenance for a digit‑recognition model, outlines integrating signing into TensorFlow/PyTorch saves, incremental hashing, GUAC for dependency graphs, and using OSV/VEX/devs.dev for vulnerability impact.
AI-written summary · about
chapters (11)
- 0:00Introduction to secure ML
- 3:55ML supply chain risks
- 7:27Model signing challenges
- 9:10Demo of model signing
- 13:28Future of model signing
- 14:50Provenance statements in ML
- 17:39Model metadata standards
- 18:43Dependency trees in ML
- 20:07Verifying model authenticity
- 21:10Vulnerability management in ML
- 21:53Q&A and resources
related talks
21mEnd-to-end secure ML Development
21mML Model Signing: Cryptographically Paving the Way to Provenance in Machine Learning Models
19mWe Know Security but How Do We Secure GenAI End-to-End?
25mFrom Model To Trust: Building Upon Tamper-proof ML Metadata Records
27mTaming the Wild West of ML: Practical Model Signing with Sigstore on Kaggle
19mModel transparency for AI/ML security
26mTaming ML’s Wild West: Practical Model Signing with Sigstore on Kaggle
26mRunning Secure AI/ML Experiment in a Unified MLOPS-DevSecOps Pipeline