Securing the ML Pipeline

What you will be able to do

• You will be able to map the end-to-end ML pipeline attack surface and threat model each trust boundary using MITRE ATLAS and the NIST AI RMF.
• You will be able to establish data and model provenance with signed lineage records and an AI bill of materials (AIBOM) that holds up under audit.
• You will be able to detect and contain malicious model artifacts, including unsafe serialization formats, before they ever reach a registry.
• You will be able to harden a model registry with signing, attestation, and least-privilege access so that only verified artifacts are promotable.
• You will be able to build a secure CI/CD pipeline for models that enforces SLSA-style provenance and policy-as-code gates at every promotion.
• You will be able to verify artifact integrity at deploy time and at runtime, and design continuous monitoring that catches tampering and drift.
• You will be able to assemble audit-ready evidence that maps your controls to the NIST AI RMF, OWASP machine learning guidance, and the NIST Secure Software Development Framework.

What is inside

4 modules, 8 lessons. Each module ends in a graded quiz and most carry an assignment.

1. 01

   Threat Modeling the ML Supply Chain

   Before you can defend the pipeline you have to see it the way an attacker does. This module maps the full ML attack surface, from training data, notebooks, and dependencies to the model registry and the serving stack, and grounds it in MITRE ATLAS and the NIST AI RMF. You will learn to draw trust boundaries and rank the supply-chain risks that actually matter for your environment.

   2 lessons · 5 quiz questions

2. 02

   Data and Model Provenance in the Supply Chain

   Integrity starts with proving where every dataset and model weight came from. This module covers data lineage, dataset integrity, and model provenance, then captures all of it in a machine-readable AI bill of materials. You will build signed provenance records that survive an audit and independent third-party review.

   2 lessons · 5 quiz questions · assignment

3. 03

   Model Registry and Artifact Integrity

   The model registry holds your production-bound models, which makes it a crown-jewel supply-chain target. This module hardens it with cryptographic signing and attestation (Sigstore, cosign, and in-toto) and teaches you to catch malicious artifacts, including unsafe pickle serialization, before promotion. You will learn to design a scan-and-quarantine workflow that lets only verified, converted, and attested models reach the trusted registry.

   2 lessons · 5 quiz questions

4. 04

   Secure CI/CD and Continuous Assurance for Models

   A model is only as trustworthy as the pipeline that built and shipped it. This module builds secure CI/CD for models with SLSA-style provenance, hermetic builds, and policy-as-code gates that block anything unverified, then extends that trust into production with deploy-time verification, runtime integrity checks, and continuous drift monitoring.

   2 lessons · 5 quiz questions · assignment