Changelog¶

All notable changes to the MLSecOps Guide are documented here.

[Unreleased]¶

GUIDE-SUMMARY.md and related summary navigation/links from README, GETTING-STARTED, MkDocs, and GitHub Pages.

references/AARM-ALIGNMENT.md — complementary mapping to CSA AARM (Autonomous Action Runtime Management); links from Ch.8 and GETTING-STARTED.

Per-section traceability (Issue #1) — References / Source mapping blocks on every major ## section across Chapters 1–17; traceability convention in Ch.15; CONTRIBUTING/GOVERNANCE expectations updated.
OWASP AI Exchange integration — complementary positioning and per-section source mapping where applicable: Ch.1 Exchange relationship; Ch.2 AI inventory + risk workflow; Ch.5 poisoning taxonomy + model theft paths; Ch.6 three test categories; Ch.7 augmentation data, downstream injection, verification gaps; Ch.12 periodic table cross-ref; GETTING-STARTED.md resource selector; Ch.15 Exchange citation.
Ch.1 — Why this guide matters — explicit reader value proposition, learning outcomes, and practical takeaways (Issue #2).
Ch.1 — Guide at a glance — audience, goal, outputs, and scope summary after Abstract.

Ch.1 — intro polish: transition sentence, split paragraphs, lifecycle lead-in, wording fixes per community review.
Ch.1–17 — framework source mapping audit: OWASP LLM/ML Top 10 IDs, MITRE ATLAS techniques, NIST AI RMF, ISO/IEC 42001, OpenSSF, and OWASP AI Exchange permalinks aligned to section topics.
GETTING-STARTED.md — Quick Start (5 Minutes), New to AI Security? path, renamed production baseline section, When to use this guide vs. OWASP AI Exchange.
TABLE-OF-CONTENTS.md — Ch.1 entries for Why this guide matters and What this guide adds; traceability convention; Exchange-related sections (Ch.1–2, 5–7, 12).
Ch.7 — removed duplicate augmentation block under Ingest security (cross-ref only).
Ch.12 / Ch.15 Appendix B — MITRE ATLAS table synchronized (AML.T0066, AML.T0034).
Ch.13 — LangChain case study: added CVE-2025-68664 alongside CVE-2025-27520.
prepare_pages.py — rewrite anchored GETTING-STARTED.md#… links for MkDocs; site index banner v1.1.0.
README.md — OWASP AI Exchange in frameworks list; version and download links v1.1.0.
Version strings aligned to v1.1.0 across Ch.1, TOC, README, CITATION.cff, CONTRIBUTING, GOVERNANCE, SECURITY, RELEASE_NOTES, and releases README.

Zenodo DOI 10.5281/zenodo.21206781 for long-term citation.
Zenodo archival release — no guide content changes; triggers DOI after GitHub–Zenodo integration.

GitHub Release v1.0.0 with DOCX and PDF (MLSecOps Practical Reference Guide).
Printable editions aligned with markdown source (16 chapters + Appendix E).
Diagram PNG assets in assets/diagrams/ (26 static Mermaid exports).

Version bump from v0.1.2 (public beta) to v1.0.0 — content scope frozen for citation.
README, TOC, and site banner aligned to v1.0.0.

GitHub Pages documentation site: https://l4tr0d3ctism.github.io/MLSecOps/ (MkDocs Material, Mermaid).
Public-beta README restructure: Getting Started, Downloads, Community feedback, Releases docs.
GETTING-STARTED.md, RELEASING.md, releases/README.md.
Chapter 16 Kubernetes deployment reference.
Appendix E: Implementation Reference — architecture cards, decision matrix, threat model template, Evidence Pack template, playbooks, master control matrix.
How to use this guide (Ch.1) and role-based Quick Start (README, docs site).
Managed AI services security reference (Ch.2) and Appendix D checklist with Evidence Pack fields.
Agent chapter priority 2: think–act cycle, conversation manipulation, exfiltration model, defense layers, secure lifecycle, KPIs.
LLM verification approach (Ch.7) complementing OWASP LLMSVS.
Positioning: MLSecOps Practical Reference Guide; What this guide adds in Ch.1 and Final Conclusion (four operational contributions).
GitHub Actions workflows: pages.yml, manual release.yml (v1.0+).

README: live documentation site link; contributor and governance docs updated.
MkDocs site built in CI from chapters-en/ (removed committed docs/ tree).
Chapter 3 condensed: demonstrated threats first; emerging topics in summary table (~450 → ~170 lines).
Ch.12 tool commands relabeled as optional appendix; Ch.12 reading note: primary mapping vs optional CLI appendix.
Removed examples/ directory; Ch.16 points to upstream vendor references only.
README, TOC, and citation text aligned with Practical Reference Guide positioning.

Unified lifecycle terminology across Ch.1, 6, 9, 11, 15, 16 (control points, release decisions, Integrity and Provenance).
Appendix order in Ch.15 (A → B → D).

Canonical policy gate model (stages 4, 7, 8) vs integrity checkpoint (stage 9); resolved minimum-baseline contradictions.
OWASP LLM04 vs LLM08 mapping for retrieval/RAG corpus poisoning (Appendix A, Claims table).
MITRE ATLAS AI worm mapping (no false 1:1 ID with memory poisoning).
Softened non-standard overclaims in conclusion; clarified OpenSSF relationship (independent, non-affiliated guide).
Case study citations and Documented incident vs Illustrative pattern labels.
Removed unverified tool reference (BlackVault); privacy/GDPR section for prompt logging; Feature Store, LoRA/PEFT, MCP sections.
Maturity Level 3 criteria (process-based, not “zero incidents”); ethics box in Chapter 3.

Initial public draft: 15 chapters, MkDocs site, threat/control mapping, maturity roadmap.