待翻译：MLSecOps Practical Guide – open-source handbook for securing AI systems

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Notifications You must be signed in to change notification settings Fork 2 Star 3 BranchesTags Open more actions menu Folders and files NameName Last commit message Last commit date Latest commit History 35 Commits 35 Commits .github .github assets/diagrams assets/diagrams chapters-en chapters-en promotion promotion references references releases releases scripts scripts seo seo .gitignore .gitignore CHANGELOG.md CHANGELOG.md CITATION.cff CITATION.cff CODE_OF_CONDUCT.md CODE_OF_CONDUCT.md CONTRIBUTING.md CONTRIBUTING.md GETTING-STARTED.md GETTING-STARTED.md GOVERNANCE.md GOVERNANCE.md LICENSE LICENSE README.md README.md RELEASE_NOTES.md RELEASE_NOTES.md RELEASING.md RELEASING.md SECURITY.md SECURITY.md mkdocs.yml mkdocs.yml requirements-pages.txt requirements-pages.txt Repository files navigation Open-source MLSecOps handbook for AI security, LLM/RAG, and secure MLOps. MLSecOps Practical Reference Guide is an open-source handbook for AI security, machine learning security, and secure MLOps across the full ML lifecycle — from data and training through deployment, runtime monitoring, SOC, and governance. It covers LLM security, RAG security, agentic AI, MCP, AI supply chain security, and DevSecOps patterns for production AI systems. Use it as a practitioner reference — not a product manual or an official OWASP, NIST, or ISO standard. Read online (recommended) mhaghighian.github.io/MLSecOps Source repository github.com/MHaghighian/MLSecOps Maintainer site mhsec.me Latest release v1.1.3 Cite (DOI) 10.5281/zenodo.21206781 Who is this for? Security engineers, ML/MLOps teams, architects, and risk owners who need a practical MLSecOps reference aligned with OWASP AI Exchange, MITRE ATLAS, NIST AI RMF, ISO/IEC 42001, OpenSSF Secure MLOps, and CSA MAESTRO — with operational controls, evidence, and rollout guidance. What this guide adds This guide synthesizes OWASP, MITRE ATLAS, NIST AI RMF, ISO/IEC 42001, OpenSSF Secure MLOps, and CSA MAESTRO. Its operational additions are: Ten lifecycle control points — one thread from change initiation through monitoring Explicit release decisions — separate evidence-producing steps from blocking gates (control points 4, 7, 8) and integrity at 9 Evidence Pack — auditable output bundle per release Implementation Reference — architecture cards, decision matrix, templates, playbooks Learn more: Chapter 1 — What this guide adds. Topics covered Area Chapters / focus MLSecOps & threat modeling Scope, risk, autonomous AI threats Data security & privacy Training data, PII, augmentation confidentiality AI / ML supply chain Model artifacts, provenance, signing Secure ML pipeline Ten control points, CI/CD gates, Evidence Pack LLM security & RAG Gateway, guardrails, prompt injection, retrieval ACL Agentic AI & MCP Tool policy, Intent Gate, scoped execution Runtime & SOC Monitoring, detection, incident response Governance & compliance Evidence, maturity roadmap, Kubernetes patterns Why MLSecOps? Traditional DevSecOps does not fully address model artifacts, training data, LLMs, RAG, agents, or runtime AI risks. MLSecOps extends existing security practices with lifecycle-specific controls, evidence generation, and AI-focused governance — without replacing your CI/CD or MLOps platform. Key features Ten-point lifecycle control model and release decision points Evidence Pack methodology per release Implementation Reference — architecture cards, templates, playbooks Threat / control / tool mapping (Ch.12) LLM, RAG, Agent, and MCP security (Ch.7 · Ch.8) AI supply chain and model artifact security (Ch.5) Kubernetes reference patterns (Ch.16) SOC integration, governance, case studies, and maturity roadmap Quick start Read online Documentation site — full guide, search, TOC Markdown Table of Contents · Chapter 1 Role-based paths GETTING-STARTED.md Contribute CONTRIBUTING.md · Issues · Discussions Role Start here Executive / risk Ch.1 → Ch.2 → Ch.14 Security engineer Ch.2 → Ch.6 → Ch.12 ML / MLOps Ch.6 → Ch.5 LLM / RAG / Agent Ch.7 → Ch.8 Production rollout Appendix E → Ch.6 Project status, roadmap, and governance: GOVERNANCE.md · CHANGELOG.md. Architecture Executive lifecycle (detail in Chapter 6): Coverage: classic ML · LLM · RAG · managed AI APIs · agents · MCP · Shadow AI · supply chain · runtime · SOC · governance · Kubernetes patterns. Downloads Latest release: v1.1.3 · Zenodo DOI Format Link Documentation site mhaghighian.github.io/MLSecOps Markdown chapters-en/ in this repository Source (ZIP) v1.1.3 archive All releases: GitHub Releases. Pre-built PDF/DOCX are not published with releases; build Word locally if you need a printable copy (below). Build DOCX yourself Generate the printable Word edition from the markdown sources: pip install -r scripts/requirements-docx.txt python scripts/build-docx.py --render-mermaid Output: dist/MLSecOps-Practical-Reference-Guide-v{version}.docx (version read from CITATION.cff). Export PDF from Word (or Pandoc) if needed. Option Purpose --render-mermaid Render missing diagram PNGs from assets/diagrams/source/*.mmd (uses system Chrome or Edge) --reference path/to/file.docx Override the Word style template --output path/to/file.docx Custom output path --skip-validate Skip post-build content checks The build uses Pandoc with the project Word template (scripts/templates/reference.docx, or auto-download from the v1.0.0 Release DOCX on first run). Template details: scripts/templates/README.md. Maintainer checklist: RELEASING.md. Frequently asked questions Question Answer What is MLSecOps? Security practices for the ML/AI lifecycle — extending DevSecOps with model, data, LLM, RAG, agent, and runtime controls. Is this an official OWASP or NIST document? No. It references those frameworks but is an independent open-source guide (CC BY-SA 4.0). Where should I start reading? Documentation site or Chapter 1. How do I cite this work? Use the Zenodo DOI or CITATION.cff. Repository structure MLSecOps/ ├── chapters-en/ # Guide chapters (English) ├── assets/diagrams/ # Diagram PNGs and Mermaid source (.mmd) ├── scripts/ # DOCX build (build-docx.py, mermaid_to_png.py) ├── dist/ # Local DOCX output (gitignored) ├── GETTING-STARTED.md # Role-based reading paths ├── CITATION.cff # Citation metadata (DOI) ├── CHANGELOG.md └── .github/workflows/ # Pages deploy, releases Community feedback We welcome review from practitioners. Bug or typo: Open an issue Suggestion / discussion: GitHub Discussions Pull request: see CONTRIBUTING.md If you review the guide and agree to be listed, we can add your name under Community reviewers (with your permission only). Share this project: linking from LinkedIn, blog posts, OWASP community threads, Dev.to, or internal security wikis helps others discover the guide and improves search visibility for MLSecOps and MLSecOps Practical Reference Guide. Cite this work See CITATION.cff for machine-readable metadata. Haghighian, M. (2026). MLSecOps Practical Reference Guide (v1.1.3). Zenodo. https://doi.org/10.5281/zenodo.21206781 Frameworks referenced OWASP AI Exchange (https://owaspai.org/) OWASP LLM Top 10 (2025) OWASP ML Top 10 (draft) OWASP Agentic / MCP MITRE ATLAS NIST AI RMF ISO/IEC 42001 · ISO/IEC 23894 EU AI Act OpenSSF MLSecOps Whitepaper CSA MAESTRO CSA AARM — AARM Alignment (agentic runtime; complementary mapping) Contributing · License CONTRIBUTING.md How to contribute LICENSE CC BY-SA 4.0 SECURITY.md Report issues in this repo CODE_OF_CONDUCT.md Community standards Questions: Issues · Discussions. Topics Resources Readme License Code of conduct Code of conduct Contributing Contributing Security policy Security policy Activity Stars 3 stars Watchers 0 watching Forks 2 forks Report repository