Resources Version 1.3 · September 2026

Securing Agentic AI in the Enterprise.

A maturity and decision framework for security, platform, and risk leaders — ten control domains, five maturity levels, five autonomy tiers, and a production gate.

Start reading → Download PDF ↓

Parts

4

Sections

25

References

32

Status

Released

The thesis

Most enterprise AI security guidance published before 2026 was written for a different technology. It assumes a model that answers questions. It does not account for a system that plans, holds credentials, invokes tools, remembers across sessions, and acts on infrastructure without a human in the loop for each step.

Controls must bound what an agent *can* do, not depend on predicting what it *will* do. Design as though the agent will attempt every action its permissions allow, because under sufficient optimization pressure it eventually will.

Written to be usable by an organization that has not yet started, and by one that has 40,000 agents it did not authorize.

Read Section 1 →

The three scales

Ten control domains, and two ordered instruments everything after Part I is scored on. The domains are a set; the levels and the tiers are scales, and they are not interchangeable.

Ten control domains

DISC

AUTH

CRED

ISO

TOOL

DATA

SUP

OBS

RESP

GOV

Five maturity levels

L0

L1

L2

L3

L4

Overall level is the minimum across domains, never the average.

Five autonomy tiers

T0

T1

T2

T3

T4

Tier attaches to a deployment, not to a product.

The 4 parts

25 sections

Scope and Models

Scope, definitions, the threat model, autonomy tiering, and the five maturity levels.

Part II §6–16

The Control Domains

The eleven control domains in detail, from discovery and identity through to retirement, oversight, and assurance.

Part III §17–23

Assurance and Adoption

The production gate, incident response, metrics, procurement, the standards crosswalk, and a sequenced roadmap.

Part IV §24–25

Appendices and References

Glossary, executive summary, and reference sources.

Contents · Securing Agentic AI in the Enterprise

v1.3

Part I — Scope and Models §1–5

01 Purpose and Scope 3 min 02 Definitions and Boundaries 2 min 03 The Threat Model 3 min 04 Autonomy Tiering: Deciding How Much Agency to Grant 2 min 05 The Maturity Model 3 min

Part II — The Control Domains §6–16

06 Discovery and Inventory 2 min 07 Identity and Authorization 2 min 08 Credentials and Secrets 2 min 09 Execution Isolation and Network Egress 2 min 10 Tool and Action Authorization 3 min 11 Data, Memory, and Context Integrity 2 min 12 Supply Chain 2 min 13 Observability and Detection 2 min 14 Incident Response and Forensics 3 min 15 Human Oversight, Governance, and Assurance 2 min 16 Decommissioning and Retirement 2 min

Part III — Assurance and Adoption §17–23

17 The Production Readiness Gate 5 min 18 Exceptions and Compensating Controls 3 min 19 Incident Response for Agent-Driven Events 5 min 20 Metrics and Board Reporting 1 min 21 Procurement Question Bank 2 min 22 Standards Crosswalk 2 min 23 Implementation Roadmap 2 min

Part IV — Appendices and References §24–25

24 Appendices 2 min 25 Reference Sources 3 min

The complete framework, typeset for print.

All 25 sections, the control domains, the production gate, the standards crosswalk, and 32 reference sources in a single PDF.

Download PDF ↓ Book advisory →