Skip to contentThe Observability LayerSearch

Enterprise handbook · Published October 2026

Agentic Responsible AI: An Enterprise Handbook

A guide to governing AI agents with clear permissions, observable outcomes, and safeguards that can be tested, written for the people building, buying, overseeing, and auditing them.

Choose a reading path

Read from the beginning or follow the decision you need to make. The glossary, worked examples, control practices, implementation specifications, and source explanations are all included.

Understand

Start with the system

Learn what an AI agent can do, whose interests it affects, and where a failure can occur.

Chapters 1–4 · Definitions, scope, governance, threats

Read the foundations
Govern

Set boundaries that hold

Define authority, protect data and memory, and give affected people meaningful review and correction.

Chapters 5–8 · Architecture, permissions, data, fairness

Explore the control design
Implement

Test the claim against reality

Build evaluations, collect outcome evidence, plan recovery, and adapt the specifications to your workflow.

Chapters 9–18 · Evaluation, operation, applications, specifications

Start the 90-day plan

Which rules does this safeguard support?

Compare all 45 handbook practices, 12 specifications and 112 reference controls with 17 selected laws, regulatory texts, standards and frameworks. Each mapping connects provisions to practical evidence, with jurisdiction, timing and applicability in view. Sources reviewed October 2026.

The complete handbook

18 chapters, 11 appendices, 14 explanatory figures, 47 core source items and 17 regulatory instruments. Each chapter opens as a readable page, with its citations and qualifications intact.

Explore the 11 reference appendices

Download the book and the implementation kit

45 proposed practices, 12 detailed implementation specifications, and 36 proposed acceptance tests connect the guidance to mechanisms, accountable owners, and observable evidence.

PDF · 123 pages · 3.0 MB · version 1.0.1

The reading edition

The complete book, including diagrams, glossary, implementation specifications, annotated bibliography, and the complete regulatory cross-tab. Tagged PDF with bookmarks. PDF/UA conformance has not been achieved.

Download the PDF
Markdown · 337 KB · version 1.0.1

The editable manuscript

Use the text in your reading and working notes. Its 14 diagrams link to their hosted files, and the implementation kit includes them too.

Download the Markdown
ZIP · 196 KB · 10 blank templates · version 1.0.1

The implementation kit

Ten blank JSON templates support the 12 implementation specifications. Use, adapt and redistribute the templates freely, including commercially, under the MIT License. The kit also includes practice and test registers, source and metric ledgers, regulatory spreadsheets, SVG diagrams, and clearly marked synthetic examples.

Download the kitHow to use the templatesTemplate reuse & license

Version 1.0.1 corrects the first edition without changing its recommendations or evidence. Read the changelog and errata. The files first published on 2 October 2026 remain available unchanged: PDF, Markdown, implementation kit.

What the evidence supports

The source notes distinguish research preprints, official legal texts, technical documentation, and voluntary frameworks. Benchmark findings retain their study conditions and denominators.

The practices and tests are proposed designs. The examples are synthetic. Neither their inclusion nor a successful download demonstrates effectiveness in your enterprise.

Inspect the annotated sources

A dated edition with monthly reference updates

The handbook covers evidence through 2 October 2026. Monthly reference updates collect newer published findings by control area. These supplementary updates do not revise the handbook text.

The complete control-family index and regulatory cross-tab are included in the handbook, alongside its 45 practices, 12 implementation specifications and proposed acceptance tests.

Follow the daily news & insights