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

The MLSecOps Practical Reference Guide, an open-source handbook for securing AI systems, has been released in version 1.1.3 by maintainer MHaghighian, providing a practitioner reference for AI security, LLM/RAG security, and secure MLOps. The guide synthesizes OWASP AI Exchange, MITRE ATLAS, NIST AI RMF, ISO/IEC 42001, OpenSSF Secure MLOps, and CSA MAESTRO, and introduces ten lifecycle control points, explicit release decisions, and an Evidence Pack for auditable output bundles. It is available on GitHub with a DOI of 10.5281/zenodo.21206781.

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 | | Source repository github.com/MHaghighian/MLSecOps https://github.com/MHaghighian/MLSecOps Maintainer site mhsec.me https://mhsec.me Latest release v1.1.3 https://github.com/MHaghighian/MLSecOps/releases/tag/v1.1.3 Cite DOI 10.5281/zenodo.21206781 https://doi.org/10.5281/zenodo.21206781 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. 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— auditable output bundle per release Evidence Pack — architecture cards, decision matrix, templates, playbooks Implementation Reference /MHaghighian/MLSecOps/blob/main/chapters-en/17-appendix-e-implementation-reference.md Learn more: Chapter 1 — What this guide adds /MHaghighian/MLSecOps/blob/main/chapters-en/01-intro.md what-this-guide-adds-beyond-owasp-openssf-and-nist . | 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 | 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. - Ten-point lifecycle control model and release decision points methodology per release Evidence Pack — architecture cards, templates, playbooks Implementation Reference /MHaghighian/MLSecOps/blob/main/chapters-en/17-appendix-e-implementation-reference.md Threat / control / tool mapping Ch.12 /MHaghighian/MLSecOps/blob/main/chapters-en/12-threat-control-tools-map.md LLM, RAG, Agent, and MCP security Ch.7 /MHaghighian/MLSecOps/blob/main/chapters-en/07-llm-rag-security.md · Ch.8 /MHaghighian/MLSecOps/blob/main/chapters-en/08-agentic-ai-security.md AI supply chain and model artifact security Ch.5 /MHaghighian/MLSecOps/blob/main/chapters-en/05-model-artifact-supply-chain.md Kubernetes reference patterns Ch.16 /MHaghighian/MLSecOps/blob/main/chapters-en/16-kubernetes-deployment-reference.md - SOC integration, governance, case studies, and maturity roadmap Read online | | Markdown Table of Contents /MHaghighian/MLSecOps/blob/main/chapters-en/TABLE-OF-CONTENTS.md · Chapter 1 /MHaghighian/MLSecOps/blob/main/chapters-en/01-intro.md Role-based paths GETTING-STARTED.md /MHaghighian/MLSecOps/blob/main/GETTING-STARTED.md Contribute CONTRIBUTING.md /MHaghighian/MLSecOps/blob/main/CONTRIBUTING.md · Issues https://github.com/MHaghighian/MLSecOps/issues · Discussions https://github.com/MHaghighian/MLSecOps/discussions | Role | Start here | |---|---| | Executive / risk | | Ch.2 /MHaghighian/MLSecOps/blob/main/chapters-en/02-scope-risk-threat-model.md → Ch.6 /MHaghighian/MLSecOps/blob/main/chapters-en/06-pipeline.md → Ch.12 /MHaghighian/MLSecOps/blob/main/chapters-en/12-threat-control-tools-map.md Ch.6 /MHaghighian/MLSecOps/blob/main/chapters-en/06-pipeline.md → Ch.5 /MHaghighian/MLSecOps/blob/main/chapters-en/05-model-artifact-supply-chain.md Ch.7 /MHaghighian/MLSecOps/blob/main/chapters-en/07-llm-rag-security.md → Ch.8 /MHaghighian/MLSecOps/blob/main/chapters-en/08-agentic-ai-security.md Appendix E /MHaghighian/MLSecOps/blob/main/chapters-en/17-appendix-e-implementation-reference.md → Ch.6 /MHaghighian/MLSecOps/blob/main/chapters-en/06-pipeline.md Project status, roadmap, and governance: GOVERNANCE.md /MHaghighian/MLSecOps/blob/main/GOVERNANCE.md · CHANGELOG.md /MHaghighian/MLSecOps/blob/main/CHANGELOG.md . Executive lifecycle detail in Chapter 6 /MHaghighian/MLSecOps/blob/main/chapters-en/06-pipeline.md : Coverage: classic ML · LLM · RAG · managed AI APIs · agents · MCP · Shadow AI · supply chain · runtime · SOC · governance · Kubernetes patterns. Latest release: v1.1.3 · Zenodo DOI https://doi.org/10.5281/zenodo.21206781 | Format | Link | |---|---| Documentation site | | Markdown chapters-en/ in this repository Source ZIP v1.1.3 archive https://github.com/MHaghighian/MLSecOps/archive/refs/tags/v1.1.3.zip All releases: GitHub Releases https://github.com/MHaghighian/MLSecOps/releases . Pre-built PDF/DOCX are not published with releases; build Word locally if you need a printable copy below . 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 /MHaghighian/MLSecOps/blob/main/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 https://github.com/MHaghighian/MLSecOps/releases/download/v1.0.0/MLSecOps-Practical-Reference-Guide-v1.0.0.docx on first run . Template details: scripts/templates/README.md /MHaghighian/MLSecOps/blob/main/scripts/templates/README.md . Maintainer checklist: RELEASING.md /MHaghighian/MLSecOps/blob/main/RELEASING.md . | 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? | | How do I cite this work? Zenodo DOI https://doi.org/10.5281/zenodo.21206781 or CITATION.cff /MHaghighian/MLSecOps/blob/main/CITATION.cff . 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 We welcome review from practitioners. Bug or typo: Open an issue https://github.com/MHaghighian/MLSecOps/issues Suggestion / discussion: GitHub Discussions https://github.com/MHaghighian/MLSecOps/discussions Pull request: see CONTRIBUTING.md /MHaghighian/MLSecOps/blob/main/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 . See CITATION.cff /MHaghighian/MLSecOps/blob/main/CITATION.cff for machine-readable metadata. Haghighian, M. 2026 . MLSecOps Practical Reference Guide v1.1.3 . Zenodo. https://doi.org/10.5281/zenodo.21206781 - OWASP AI Exchange https://owaspai.org/ 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 /MHaghighian/MLSecOps/blob/main/references/AARM-ALIGNMENT.md agentic runtime; complementary mapping | LICENSE /MHaghighian/MLSecOps/blob/main/LICENSE SECURITY.md /MHaghighian/MLSecOps/blob/main/SECURITY.md CODE OF CONDUCT.md /MHaghighian/MLSecOps/blob/main/CODE OF CONDUCT.md Questions: Issues https://github.com/MHaghighian/MLSecOps/issues · Discussions https://github.com/MHaghighian/MLSecOps/discussions .