cd /news/ai-safety/mlsecops-practical-guide-open-source… · home topics ai-safety article
[ARTICLE · art-96211] src=github.com ↗ pub= topic=ai-safety verified=true sentiment=· neutral

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.

read4 min views2 publishedAug 14, 2026
MLSecOps Practical Guide – open-source handbook for securing AI systems
Image: source

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 repositorygithub.com/MHaghighian/MLSecOps** Maintainer site**mhsec.me** Latest release**v1.1.3** Cite (DOI)**10.5281/zenodo.21206781Security 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 releaseEvidence Pack

— architecture cards, decision matrix, templates, playbooksImplementation Reference

Learn more: Chapter 1 — What this guide adds.

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 releaseEvidence Pack

— architecture cards, templates, playbooksImplementation ReferenceThreat / 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

Read online | |

MarkdownTable of Contents·Chapter 1Role-based pathsGETTING-STARTED.md** Contribute**CONTRIBUTING.md·Issues·Discussions| Role | Start here | |---|---| | Executive / risk | |

Ch.2Ch.6Ch.12Ch.6Ch.5Ch.7Ch.8Appendix ECh.6Project status, roadmap, and governance: GOVERNANCE.md · CHANGELOG.md.

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.

Latest release: v1.1.3 · Zenodo DOI

Format Link
Documentation site

Markdownchapters-en/

in this repositorySource (ZIP)v1.1.3 archiveAll releases: GitHub 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). 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.

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 DOIorCITATION.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** Suggestion / discussion:GitHub Discussions Pull request:**seeCONTRIBUTING.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 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/) - 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)

|

LICENSESECURITY.mdCODE_OF_CONDUCT.mdQuestions: Issues · Discussions.

── more in #ai-safety 4 stories · sorted by recency
── more on @mhaghighian 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/mlsecops-practical-g…] indexed:0 read:4min 2026-08-14 ·