
Agentic AI & runtime safety · Published · 2026
Runtime Safety for Multi-Agent AI Systems.
Principles, Patterns, and Practices for Safe and Reliable Autonomous Collaboration
A perfectly aligned model can still approve a fraudulent payment. The model behaves as trained; the system fails, because nothing between the model's decision and the world checks what the action will actually do.
This book treats runtime governance as an engineering discipline for multi-agent AI: pre-execution validation, in-execution monitoring, post-execution containment, shared approval under concurrency, safe delegation and omission-aware evidence, built around one running example and closing with a deployment blueprint.
For researchers and engineering teams interested in bounded, reviewable agent coordination.
Read, order & free sample
Read the full book (Google Drive) Order this bookSold directly by the author. Send an order request and I will reply by email to confirm the price, format and delivery. Nothing is charged on this site.
Read a free sample (PDF, 17 pages) Ask about this bookMetadata and imprint supplied by the owner. The free sample is taken from the book edition (EPUB, August 2026). No ISBN or price is confirmed yet; orders are taken directly on this site.
FREE SAMPLE
Read the first chapter
before you buy.
The free sample is a 17-page PDF with the title page, the table of contents and Chapter 1: From Aligned Models to Governed Agent Systems. It is an excerpt only; the complete book can be read free on Google Drive or ordered directly from the author.
THEMATIC THREADS
01Principles for runtime safety
A theme of the book. The free sample includes the table of contents and the first chapter.
02Patterns for autonomous collaboration
A theme of the book. The free sample includes the table of contents and the first chapter.
03Practices for reliable systems
A theme of the book. The free sample includes the table of contents and the first chapter.
KEEP READING


