SETTING GLOBAL STANDARDS FOR TRUSTED AI CREDENTIALSAI Competence Framework v1.29 · current release
You are reading the current version of the framework, v1.29.Permanent address for this version

How the framework is put together

The framework is a stack of six layers and three axes. This page makes both concrete, and states the rule that keeps one specification able to support any number of credentials.

B.1 — The layer model

Six layers. Each references the layer above. No layer redefines what a higher layer owns.

LayerNameOwned byContains
0MethodologyAI Certification StandardsHow statements are written, reviewed and versioned
1Core frameworkAI Certification StandardsDomains D1–D8, levels L1–L4, competence statements
2OverlaysAI Certification StandardsStatements specific to a professional function, applied across domains
3ExtensionsAI Certification Standards, on submissionStatements specific to a technology, sector or business function
4Occupational standardsAI Certification StandardsA named role at a named level, expressed as a set of statements
5Credential specificationsAwarding bodiesWhich statements are assessed, how, and to what standard
6Courses and materialsAny providerDelivery, sequence, content
The governing rule

Layers 4, 5 and 6 select and weight statements. They never create one. Where a credential requires a competence the framework does not contain, an extension is proposed at layer 3 and enters public review. This is what allows one framework to support any number of credentials without fragmenting.

B.2 — The three axes

Domain, level, context

Every statement sits at the intersection of a domain (what area of competence), a level (to what degree of autonomy and judgement), and optionally an overlay or extension (in what professional or applied context).

D3.L2.04Interaction and elicitation, Applied level, fourth statement
O1.L3.02Instruction and assessment overlay, Proficient level
X-AGT.L2.01Agentic systems extension, Applied level
B.3 — Domains, overlays and extensions distinguished

The distinction is the mechanism by which the framework grows without becoming unmanageable, and it is worth stating precisely.

UnitAnswersApplies toExample
DomainWhat area of competence?Everyone, in any roleD8 Security
OverlayWhat professional function?Those performing that functionO1 Instruction and assessment
ExtensionWhat applied context?Those working in that contextX-AGT Agentic systems

A marketing manager using AI is not performing a different professional function from an analyst — they are working in a different context. That is an extension. An instructor teaching AI competence is performing a distinct function that carries its own obligations regardless of context. That is an overlay.

B.4 — How a credential is assembled

Every credential has the same form

Credential = { core statements } ∪ { overlay statements } ∪ { extension statements }
             each with a coverage flag and a weighting

Because the identifiers are shared, the overlap between this credential and one for agentic governance is visible and measurable. A holder of one can be granted recognition for the statements common to both without either awarding body negotiating a bilateral agreement.

Worked example — security practitioners working with agentic systems
ComponentStatementsWeight
CoreD8 at Proficient45%
CoreD6 at Applied15%
ExtensionX-AGT security-tagged, Proficient30%
OverlayO3 at Applied10%
B.5 — How to read a statement

Every field, annotated against a real statement

D6.L3.06

Critique a published benchmark or vendor claim, identifying what it does and does not establish.

Indicators · normative
Identifies what was measured, on what data, under what conditions
States the gap between what was measured and the intended use
Distinguishes an unsupported claim from a supported claim about something else
Assumes D6.L1.04 · Related D1.L3.*, D7.L3.*
IdentifierPermanent. Never reused, renumbered or removed. The address of the statement is the identifier.
StatementOne observable competence at one level, opening with a verb from that level’s controlled list.
TypeKnowledge, Skill, Judgement or Practice. Determines which assessment methods can reach it.
IndicatorsObservable behaviours evidencing that the statement is met. Normative.
Evidence examplesIllustrations of how the statement might be assessed. Non-normative.
AssumesPrerequisite statements, resolved as links.
RelatedCross-references, including across domains, with the relationship stated.
StatusPublished, in draft or deprecated, with the version introduced and last modified.
StabilityWhether the meaning may change at a minor release.

This page displays version 1.29 · last reviewed 30.08.2026