Few organizations train their own foundation models. They download weights from model hubs, fine-tune with open datasets, install fast-moving ML libraries, pull inference containers and connect agent tools and MCP servers written by others. Every one of those links can be tampered with, and some, such as pickle-based model files, can execute code the moment they are loaded. This course teaches you to secure that chain with the same rigor mature teams apply to software, using frameworks current as of October 2026: SLSA v1.2 (build and source tracks), NIST SP 800-218 and its generative AI profile SP 800-218A, the OpenSSF Model Signing (OMS) specification with Sigstore, CycloneDX ML-BOM and the SPDX 3.0 AI and Dataset profiles.
You will learn why safetensors and similar formats reduce risk, what model scanners can and cannot detect, and why backdoored weights are hard to find by inspection. You will sign and verify model artifacts, capture provenance for training and fine-tuning pipelines, enforce verification at deployment with admission policies, and maintain AI BOMs, model cards and license records through retraining. The course covers ML framework and inference server vulnerabilities, third-party model APIs, and agent tools and MCP servers as supply chain, then closes with a model registry and promotion gates, incident response for a compromised artifact, and a maturity roadmap. The Wirewalk Institute certificate attests to completion and assessment of this course; it is not endorsed by OWASP, MITRE, NIST or any vendor named here.