ai · security · skills

Sample consultancy report

Globex: Application & DevSecAIOps Security

Where AI has moved into this function, whether the controls kept up, and the one move that earns the next rung. Read in five minutes; decided in one meeting.

Prepared on
The aisecurityskills function diagnostic
Basis
Cloud Security Alliance (CSA) AI Controls Matrix (AICM) · AI Security Maturity Model (AISMM) × AI Cyber Maturity Model (AI-CMM)
Evidence state
Illustrative · fictional organisation
Issued
18 Jul 2026
01

The verdict

Governance ahead of adoption: AppSec policy and testing exist, but agent boundaries and protection-by-design still cap the climb while most build workflows stay human-run.

Globex's application and DevSecAIOps security function reads at L2.2 governance against L1.2 autonomy: an industrial posture where the gate is open because adoption has not outrun the evidence. Identity and access foundations reach Defined; AppSec policy and testing are in place. What holds the composite down is the absent agent-boundary control, absent protection-by-design on AI data pipelines, and named supervision that is only partial. Code review is assisted; dependency scanning, threat modeling, AppSec testing, and model-build assurance remain largely manual — governance must close those gaps before autonomy claims another rung.

Governance

L2.2

AI Security Maturity Model (AISMM) · how well it is secured

Autonomy

L1.2

AI Cyber Maturity Model (AI-CMM) · how far AI has gone

The gate

Open

margin +0.8 · autonomy inside what governance allows

02

Where this function stands

The whole method in one drawing. Governance runs across, autonomy runs up, and the staircase is the gate: each governance level earned is the autonomy an organisation may responsibly claim.

Globex on the governance-and-autonomy gridGovernance from L1 to L5 across, autonomy from L1 to L4 up. The gate staircase marks the autonomy each governance level has earned. Globex sits at governance 2.2, autonomy 1.2, inside the governed region.Ungoverned: ahead of controlsGoverned: earned autonomyGlobex today · L2.2 / L1.2L1L2L3L4L5L1L2L3L4Governance (AISMM) → each level earned is autonomy allowedAutonomy (AI-CMM) ↑
The gate is the product’s one rule: autonomy must never outrun governance. This function sits at L2.2 governance, which allows autonomy up to L2 — and it runs at L1.2, inside the line. Climbing the wall, not the drop.
03

What was measured

Two ladders, one instrument. Each answer maps to a Cloud Security Alliance AI Controls Matrix (AICM) control; a category claims a level only when that tier is evidenced, and an absent control caps it. The same rule scores every live run.

How well it is secured

L2.2

InitialRepeatableDefinedCapableEfficient

The CSA AI Security Maturity Model (AISMM): every answer maps to an AI Controls Matrix (AICM) control, and a level is claimed only when the tier is evidenced.

How far AI has gone

L1.2

ManualAssistedAugmentedAutonomous

The AI Cyber Maturity Model (AI-CMM): where the human sits in each workflow — in, on, then over the loop. our model · calibrated to SAE J3016.

IAML3 Defined · 2 of 3 evidenced

Held below the next tier: tier 4 not yet evidenced

Infrastructure Security and ResilienceL2 Repeatable · 1 of 3 evidenced

Held below the next tier: tier 3 not yet evidenced

Data SecurityL2 Repeatable · 1 of 2 evidenced

Held below the next tier: DSP-07 absent

App SecurityL2 Repeatable · 2 of 4 evidenced

Held below the next tier: AIS-11 absent

GovernanceL2 Repeatable · 1 of 2 evidenced

Held below the next tier: tier 3 not yet evidenced

04

Findings

Three, ranked, classified by what leadership does with each: act on a priority, protect a strength, and hold a deliberate choice.

  1. 01Priority

    Agent boundaries and protection-by-design are still absent

    App Security caps at Repeatable because agent tool, data, and action boundaries are not defined, and Data Security caps because protection-by-design is absent on AI pipelines. Those two absents are the board priority: without them, Defined evidence in AppSec and data cannot be claimed even though policy and testing already run.

    The exact control ids (for your security and governance, risk and compliance team)

    AIS-11 · DSP-07 · AIS-05

  2. 02Strength

    Identity foundations reach Defined while AppSec policy is written

    An identity and access policy covers AI systems, least privilege is enforced on models and pipelines, and an application security policy plus testing cover AI-enabled apps. Classification of AI data flows is implemented underneath. That floor is what keeps governance ahead of the still-manual adopt average.

    The exact control ids (for your security and governance, risk and compliance team)

    IAM-01 · IAM-05 · AIS-01 · DSP-04

  3. 03By design

    Most build workflows stay human while governance finishes Defined

    Code review runs assisted; dependency scanning, threat modeling, AppSec testing, and model-build assurance stay manual on purpose. Engineers are trained on acceptable use, but named human supervision of AI-assisted engineering is only partial. That restraint is what keeps the gate open with positive margin.

    The exact control ids (for your security and governance, risk and compliance team)

    HRS-15 · GRC-15 · AIS-03

05

The climb

Direction, not a how-to: the next rung, and the governance that must move before autonomy does.

  1. Next quarter

    Define agent security boundaries (tools, data reach, allowed actions) and apply protection-by-design to AI data pipelines so AppSec and Data Security can evidence Defined.

  2. Two quarters

    Name accountable supervision for AI-assisted engineering work and close the access-review and AppSec-metrics partials so Capable evidence matches the autonomy the team is ready to claim on code review.

  3. Continuous

    Hold AppSec testing and least-privilege enforcement through each model and pipeline change; re-run the diagnostic before lifting dependency or model-build workflows another autonomy rung.

06

About this instrument

What a reader should carry out of the room: how the diagnostic works, how progress is tracked, and what the practice is for.

One questionnaire, two reads

Every answer maps to a Cloud Security Alliance AI Controls Matrix (AICM) control. Read one way, the answers grade the function: governance versus autonomy, joined by the gate. Read the other way, the same answers name the skills each person in the function must acquire. Diagnosis and reskilling from one sitting.

Tracked, not judged

The first run is a baseline, never a verdict. Re-assess after the work and the radar overlays the previous run, so leadership sees movement, not a grade. The compatible-standard packs (ISO/IEC, the National Institute of Standards and Technology, and the CSA AI Consensus Assessments Initiative Questionnaire) are lenses on the same answers: assess once, report many ways.

Direction, not a solution

AI is a moving target, so the report names the next rung and the governance that must move first — never a vendor stack or a how-to. The gate keeps the climb honest: autonomy is claimed only after the controls that catch it are in place.

Derived at build time from the Globex posture config through the live function-diagnostic scorer: the same questions, tiers, and gate every real run uses. A bank change re-derives this sample automatically; nothing here is hand-scored.

Globex is a fictional organisation; the postures are self-assessed sample data, never client results. Nothing in this report is certification, and no standards body has reviewed it.