Before the Tool Call
Chapter 10 promised a firewall for agent reasoning. One hundred and eighty-nine days later, here it is, and here the book ends.
Who Holds the Undo?
Six cyber agencies said to classify agent actions by reversibility. None drew the matrix.
The Vendor's Test Was the Threat
No law required OpenAI to disclose the breach to Australia. It took 84 days anyway.
The Control Existed. The Evidence Didn't.
MBIE and Health NZ each had governance on paper. Neither had independent proof it was working until someone else went looking for it.
AI Governance: The Exception Log Boards Should Be Asking For
A policy attestation tells a board a document exists, not whether anyone followed it.
Zero Trust for AI Agents: The Control That Stops It Going In Is Not the Control That Stops It Coming Back Out
OWASP already splits ingress from egress. Most control estates report only one number.
A Firewall for a Protocol Nobody Told You Was on the Network
Microsoft's MCP firewall finds shadow servers by deep packet inspection, not a catalogue
Your Experiment Tracker Is Production Infrastructure Now
CISA gave Ray three days to patch and MLflow fourteen. The gap is the assessment.
The Trust Boundary Is the CI Runner
AIShellJack's 84% attack rate and three 2026 disclosures point to one design gap
Ninety-One Per Cent of Your Identities Are Not People: Zero Trust for the Machine Majority
A new measurement shows most of what authenticates in production was never human
Governance Without Architecture Is Incomplete: The Security Manager Becomes an Architect
ISACA rebuilt the CISM exam around architecture fluency, not controls alone
The Same Model, Two Regimes: What 10 December 2026 Requires of Trans-Tasman AI Architecture
One deadline, two jurisdictions, and the automated-decision inventory most AI governance roadmaps skip
The Delegation Problem
Why 40 Per Cent of Enterprises Are Heading for a Post-Incident Reckoning with OAuth 2.1
Every Agent Is a Login: FIDO2, Passkeys, and the Human-Facing Identity Boundary
When phishing-resistant authentication meets agentic AI, the question is not whether your agents can authenticate. It is whether the human who authorised them was actually there.
When Five Governments Tell You to Slow Down
SPIFFE, SVIDs, and the Architecture of Cryptographic Agent Trust
The Non-Human Identity Crisis: Why Your Zero Trust Architecture Has a Blind Spot
Part Four: Identity Architecture | Chapter 16
The Non-Human Identity Crisis: Why Your Zero Trust Architecture Has a Blind Spot
Part Four: Identity Architecture | Chapter [TBC; recommended: 19]
When Insurers Become the Regulator: The Cyber Insurance Path to Zero Trust
Your renewal questionnaire is now part of your architecture roadmap. Here is what that means for NZ practitioners.
Why Your AI Architecture Must Understand Māori Data Sovereignty
The six principles your Zero Trust framework doesn't mention — and why they govern more than you think