Methodology

A structured view of AI governance maturity.

SHIELD translates recognised obligations and management practices into six areas leaders can examine, evidence and improve.

Regulatory mapping

5 baselines, mapped without false equivalence.

Each baseline has a different legal status and scope. SHIELD reports each one separately rather than combining them into a single legal conclusion.

European Union

EU AI Act

Risk-tiered obligations for providers and deployers of AI systems touching the EU market.

Sets the classification, documentation, transparency and human-oversight duties that other regimes are converging on.

ISO/IEC

ISO/IEC 42001

Certifiable management-system requirements for governing AI across its lifecycle.

The practical checklist an auditor uses: policy, roles, impact assessment, controls, review.

Hong Kong regulators

SFC / HKMA guidance

Expectations for licensed and authorised institutions using AI in regulated activity.

Defines senior-management accountability, model validation and outsourcing duties in this market.

US NIST

NIST AI RMF 1.0

Voluntary framework organised around Govern, Map, Measure and Manage.

Gives a common language for describing AI risk to boards, clients and counterparties.

Cyberspace Administration of China and sector regulators

Mainland China

Filing, labelling, security assessment and data-export duties for AI services offered in or from Mainland China.

Adds algorithm filing, content labelling and cross-border data controls that the other regimes do not require.

Scoring version 1.1 (gap ranking v2)

How this is scored

The assessment has 24 questions, four per SHIELD letter. Each answer scores 0, 1, 2 or 3.

A dimension score is its total divided by its maximum score of 12, shown as a percentage. The overall score is the average of the six dimension scores. All six dimensions carry equal weight.

Each question is tagged to the frameworks it relates to. A framework status comes from the average of its tagged answers: On track at 2 or above, Partial at 1 or above, and Gap below 1. Every respondent currently sees all 5 frameworks, whether or not they apply. The context questions do not change the score yet.

Any answer of 0 or 1 becomes a gap. Gaps are ranked by the lowest answer first. Where answers are equal, gaps are ordered by dimension: D, L, H, I, S, then E. The number of frameworks a question touches never raises its priority.

SHIELD score bands
BandScore
Leading75 and above
Progressing50 to 74
Emerging25 to 49
ExposedBelow 25

These bands are SHIELD conventions, not validated risk thresholds.

This is a self-assessment completed by one person. It is an AI governance maturity metric, not legal advice, a certification or a compliance determination. Results from this version will not be directly comparable with future versions.