AI Certification Standards — articles Non-normative explanation and commentary, bylined and revisable. https://aicertificationstandards.org/articles.xml 2026-08-28T00:00:00Z A benchmark result is not a fitness claim https://aicertificationstandards.org/articles/a-benchmark-result-is-not-a-fitness-claim 2026-08-28T00:00:00Z A benchmark measures performance on the benchmark. Treating that number as evidence of fitness for a particular task is the most common assurance failure in current practice. What “aligned to the framework” is allowed to mean https://aicertificationstandards.org/articles/what-aligned-is-allowed-to-mean 2026-08-27T00:00:00Z The four tiers in plain language, and the exact wording each one permits. Tier 0 is a declaration; only Tiers I and II involve us reading anything. Telling L2 from L3 https://aicertificationstandards.org/articles/telling-l2-from-l3 2026-08-29T00:00:00Z Two tests that settle most level disputes: who chose the approach, and what happens when the situation is not the one the procedure anticipated. Why security is a domain and not a lens https://aicertificationstandards.org/articles/why-security-is-a-domain 2026-08-25T00:00:00Z D8 exists as a domain because a lens applied across the other seven domains would have been invisible in every coverage map. The choice has real costs, stated here. What a coverage map claims, and what it does not https://aicertificationstandards.org/articles/what-a-coverage-map-claims 2026-08-24T00:00:00Z Taught and assessed are different claims. The gap between them is where a coverage map is most often used to imply something it does not say. Why statement identifiers are never reused https://aicertificationstandards.org/articles/why-identifiers-are-never-reused 2026-08-23T00:00:00Z An identifier is a contract with every credential that cited it. Renumbering would silently change what those credentials claim, years after they were awarded. Verifying AI output is a competence, and almost nobody teaches it https://aicertificationstandards.org/articles/verifying-output-is-a-competence 2026-08-22T00:00:00Z Courses teach people to produce output and stop there. The competence that determines whether the output is usable is left to be picked up on the job, if at all. Why one framework and not one per certification https://aicertificationstandards.org/articles/why-one-framework 2026-08-21T00:00:00Z Every awarding body could publish its own competence framework, and several have. The layer model exists because per-credential frameworks fragment the thing they claim to standardise.