SETTING GLOBAL STANDARDS FOR TRUSTED AI CREDENTIALSAI Competence Framework v1.29 · current release

Articles

Explanation, reasoning and worked practice, bylined and dated. Written by the people who made the decisions being explained, because nobody outside this organisation can write those pieces.

Nothing here is part of the specification. Articles are bylined, may be revised, and carry no normative weight. The formal record — releases, consultations, conformance decisions, recusals — is at Notices, and is never merged with this stream.

8 articles published · 6 commissioned and unwritten · 4 authors, each with declared interests · non-normative throughout

Filter articles

/articles

8 articles, most recent first · no article is featured or promoted

Telling L2 from L3

Two tests that settle most level disputes: who chose the approach, and what happens when the situation is not the one the procedure anticipated.

D6 — D6.L2.01 · D6.L2.02 · D6.L3.01 · D6.L3.04
Five types

What each type is, and which carry a stronger warning

ExplainerWhy the framework is built as it is. Structural reasoning.4 published
Guidance noteHow to do something. Non-normative, but practical and specific.2 published
CommentaryA view on practice or on the field. Clearly opinion.2 published
Case studyAn account of applying the framework, with the organisation's consent.0 published
ReviewAssessment of a tool, claim, benchmark or standard.0 published

Commentary and Review carry the stronger banner because they are argument rather than explanation. Type tags are drawn in steel, not in the status colours: these are categories, not states.

In preparation

6 articles commissioned and not yet written

These are commissioned and unwritten. They are listed rather than published as stubs, because an article that does not exist is not a page.

ExplainerThe fifth level, and why it is empty

What reserving a level protects against.

ExplainerExtensions, and how a framework grows without breaking

Why agent security is an extension rather than a ninth domain.

ExplainerWhy the framework contains no product names

Tool neutrality and its cost.

Guidance noteReading a credential’s coverage map before you buy

For individuals, in the order the questions actually arise.

Guidance noteWriting a job description against the framework

For employers, with the two mistakes that make a role unfillable.

CommentaryUnsanctioned AI use is a governance failure before it is a user failure

From the D8 editorial notes.

Corrections and revisions

Articles may be updated. Notices may not.

A substantive change appends a revision entry with its date and a note saying what changed. The revision history renders on the article.
A factual correction is marked in the body at the point of correction, not silently fixed. The original claim stays visible with the correction beside it.
A withdrawn article stays at its address with a withdrawal notice and the reason. It is not deleted, and it is not quietly unlinked.
A superseded article stays, marked, linking to what superseded it. The formal record at Notices works differently: a notice is never edited at all.

Today: 1 article revised since publication, no article withdrawn, no article superseded.

Authors

Everyone who writes here, with their interests

Björn HaglundRelated party1 articles

Founder and Director, AI Certification Standards

Sofia Lindqvist3 articles

Editorial lead, D6 Evaluation and assurance

Rasmus Dahl1 articles

Editorial lead, D8 Security

Registrar, AI Certification Standards

Page last reviewed 30.08.2026