Insights
Essays on the AI-nativesecurity practice.
Anchored in the methodology and case studies documented on this site. Published as the substance is ready, not on a schedule.
Featured case study · modeled scenario
Transforming a Security Operations Center.
The end-to-end worked example: the trap most SOCs fall into, what changes when analysts get the right skills, and the five-phase transformation, phase by phase, with named skills.
Read the case study →
All essays, newest first
9 July 2026 · 16 min read
Choosing the Control Spine: CSA, Forrester AEGIS, and Gartner AI TRiSM
The AI-security framework landscape read as a practitioner: what each artifact actually is, why this practice anchors on the Cloud Security Alliance (CSA) AI Controls Matrix and AI Security Maturity Model, and how the analyst lenses are integrated rather than ignored.
Three kinds of artifact compete for the same decision: open consensus control standards, paywalled analyst operating frameworks, and management-system standards written for auditors. Why the spine of this practice is CSA’s AICM (18 domains) and AISMM (12 categories), where Forrester’s AEGIS and Gartner’s AI TRiSM genuinely add what CSA lacks, and how all of it lands in one app: the 8-function ownership lens, the maturity radar, the gate, and the loop.
Read the essay →
11 June 2026 · 6 min read
What Claude Mythos Means for Your Security Program
Thousands of AI-discovered zero-days, exploit chaining, autonomous attack capability — the 2026 coverage every Chief Information Security Officer (CISO) read, mapped to what concretely changes in your program.
Anthropic’s Mythos found thousands of zero-days and chained them into exploit paths; the UK AI Security Institute (AISI) showed autonomous attack capability against weakly defended systems. The honest reading: the vulnerabilities were always there — speed, volume and autonomy changed. Four challenges from the public coverage, and the framework answer to each: the vulnerability-prioritization play, the autonomy gate, the Inheritance Stack, and the People dimension no tooling can substitute.
Read the essay →
9 June 2026 · 7 min read
How a Skill Is Built and Fitted to Your Function
The two halves of the engine: how a skill is authored, step by step, and how the raw library is configured to a function’s deployed tools and real process via its fit-points.
A thousand unfitted skills are inventory, not capability. How a skill is structured as an executable procedure, the seven steps to author one, and the six kinds of fit-point that bind a generic skill to a function’s tools and process — the configuration that turns raw material into a measured result.
Read the essay →
3 June 2026 · 5 min read
The AICM Coverage Assessment
Step-one deliverable: a YES / NO / NA answer for every control objective — each one earned with evidence, never self-asserted.
A maturity grade without a control inventory is an opinion. The coverage assessment fixes the territory first: one evidence-backed answer per AI Controls Matrix (AICM) control objective, partials made explicit, gaps written down — the defensible baseline everything else builds on.
Read the essay →
3 June 2026 · 5 min read
Placing a Function on the AISMM Scale
Step-one deliverable: a specific Initial → Efficient level per AI Security Maturity Model (AISMM) category, with the evidence and reasoning that justify it.
The scorecard grades the checklist, not the other way round. How a maturity placement is bounded by coverage evidence, justified in writing, and moved one defensible rung at a time — and why a level you cannot re-derive is a mood, not a level.
Read the essay →
3 June 2026 · 5 min read
The Shared Security Responsibility Model (SSRM) Ownership Map
Step-one deliverable: one named owner for every in-scope control across your AI supply chain — client, app, orchestrator, model, cloud.
AI controls fail quietly in the gaps between organisations. The shared-responsibility matrix surfaces ownership gaps, chain mismatches, and the justified NAs that become contract clauses — the list of controls you believed someone else was doing.
Read the essay →
3 June 2026 · 4 min read
The Prioritised Remediation Roadmap
Step-one deliverable: corrective action per gap — owner, milestones, due date — ranked by effort-to-impact.
Severity-ordered backlogs stall; effort-to-impact ships defensible wins early and funds the harder lines. The roadmap is the budget conversation pre-written: what it costs, what it moves, who is on the hook, and what artefact flips the answer to YES.
Read the essay →
3 June 2026 · 5 min read
The External-Assurance Artefact
Step-one deliverable: the AI Consensus Assessments Initiative Questionnaire (AI-CAIQ), pointed both ways — a Security, Trust, Assurance and Risk (STAR) for AI Level 1 submission for providers, a provider-evaluation pack for AI customers.
The same questionnaire serves two audiences on the same control spine. What the AI-CAIQ attests, what STAR for AI Level 1 is — and emphatically is not — and how the provider-evaluation pack turns supply-chain due diligence into a question set.
Read the essay →
31 May 2026 · 12 min read
Operating the Platform: A Leverage Guide
The companion to the build playbook — not what we build, but how to operate it on this platform to extract the most value, by role and by move.
The platform is the operating system for our AI Security & Safety Center of Excellence (CoE): a vetted skill substrate, the AI Controls Matrix (AICM) control spine that organises it, and a deployment method that turns it into measured capability. This is how to run it — the four moves, the three lenses, the skill as a unit of work, and a per-role quick-start.
Read the essay →
31 May 2026 · 35 min read
The AI Security & Safety Center of Excellence: A Modular Build Playbook
An internal-build playbook anchored on the Cloud Security Alliance (CSA) AI Controls Matrix — pick-and-choose modules to assemble the case and map the skill library onto the control spine.
Build an internal AI Security & Safety Center of Excellence (CoE) as a hub-and-spoke operating model anchored on the CSA AI Controls Matrix (18 domains, 243 control objectives) as the master control spine — sequenced from the Security Operations Center (SOC), measured honestly, governed by control coverage and autonomy gates.
Read the essay →
26 May 2026 · 14 min read
Why Most AI Security Pilots Don't Survive Production
Five structural failure modes — and what protects the pilots that scale.
Between 2023 and 2025, the global enterprise market spent four to seven billion dollars on AI security pilots. Most produced no production capability. The five failure modes responsible are observable, structural, and avoidable — if named early enough.
Read the essay →