ai · security · skills

Perspective · for the CISO

Stand up an AI Security CoE, without it becoming shelf-ware.

Every function is adopting AI on its own terms, and “buy a platform” doesn’t build the capability or the culture. An AI Security & Safety Center of Excellence centralises the governance, prevents Shadow AI, and builds the AI-first way of working — if it’s structured to deliver, not to produce decks.

SCATTEREDper-function pilots, no shared spine → Shadow AIHUB-AND-SPOKECoEthe hubhub sets governance · function spokes run it

A federated model: one control spine and one culture at the centre, delivery owned at the edge.

The trap to avoid

A CoE that owns everything becomes a bottleneck; one that owns nothing becomes a newsletter. The build has to centralise the spine and the standard while pushing delivery to the functions.

“How do I build an AI Security CoE that actually ships capability — and then operate it?”

The answer

Anchor the whole CoE on one control spine.

The CoE is built on the CSA AI Controls Matrix — 247 objectives across 18 domains — as the master spine every skill, assessment, and maturity claim attaches to. Sequenced from the SOC, measured honestly, governed by control coverage and the autonomy gate. Because ~85% of AICM controls extend existing cloud controls, this is faster than it looks.

This document is what you build and why. For how to operate it — by move, by lens, and by role — see the companion operating guide.

How the practice operates, end to end →

The assembly kit

Twelve modules: assemble the case you need.

Pick and choose to build the pitch for your board, your functions, and your auditors. Each module is a self-contained piece of the case.

01Charter & mandate
02The strategic case (build, not buy)
03Operating model & org design
04CSA AICM control architecture
05Maturity model
06AI-by-design / human-AI division of labour
07Threat landscape & why now
08Rollout & sequencing
09Measurement & KPIs
10Stakeholder value propositions
11Service / skill-library mapping
12Governance, risk & responsible AI
Start with a function baseline →
← Back to Insights

Subscribe for the next essay.