Method, sources and versions

How the atlas is built, what each concept stands for, and the version history.

The model

The base object is always the AI system (sometimes a model, a product or the organisation). Each regulation defines qualification axes. The verdict on those axes, combined with the organisation role, makes requirements applicable. Each requirement is filed under one or more control themes and proven by checks. A check is shared by every regulation that uses it: that is what produces the crosswalk.

Jurisdiction

Territory issuing the text (EU, US state, country). Carries the region and the supervisory authority.

Framework / regulation

Text or standard: kind, status, bindingness, penalties, sources.

Milestone

Application date, certain or potential, with the obligation it triggers.

AI system

Base object of every analysis. A model, product or organisation can also be the object (GPAI, CRA, ISO 42001).

Qualification axis

Verdict palette (ordinal or categorical): AI Act risk level, role, GPAI status, Colorado consequential decision…

Criterion

Sub-qualification grouped by category (Art. 5, Annex III, responsible AI risk category).

Verdict

Dated result, proposed then confirmed, with provenance (agent, workflow, human) and history chain.

Applicability rule

Condition making a requirement applicable: verdict × role × threshold × sector × territory.

Requirement

Obligation derived from an article. Instantiated per system when applicable.

Control theme

Cross-cutting family (24 themes) linking requirements of every text: the crosswalk pivot.

Check

Evidence-gathering action, with scope (organisation, system, provider, model) and recurrence. Shared across referentials.

Evidence

Validated document, link or text, with expiry. One piece of evidence answers every requirement linked to the check.

Crosswalk

Coverage of the other referentials by the checks already live in the plan.

Date certainty

Each milestone carries a certainty level. « Set in the text »: a date written in an adopted text. « Potential »: a pending proposal, an announced delay or an estimate. « To verify »: a date that may have changed since the last check.

PastSet in the textPotentialTo verify

Sources and verification

The EU AI Act, NIST AI RMF and Brazil AI Bill requirements come from the TrustFlow referential, built with the OYAT law firm. The other sheets are written from the official texts cited on each page. Data checked on Sep 25, 2026.

General information, not legal advice. Check the official texts and get advice for your situation.

Data as JSON (all rights reserved, reuse requires written permission): atlas-0.1.0.json · atlas-latest.json

Versions

MAJOR: a public URL or the JSON format changes. MINOR: a framework, theme or section is added. PATCH: dates, articles, wording.

v0.1.0 Sep 28, 2026

First public preview.

  • 23 frameworks across 15 jurisdictions, 231 requirements, 144 checks (19 proposed).
  • One fact sheet per framework in French and English, 24 control theme pages, crosswalk, timeline, responsible AI taxonomy, method page.
  • Versioned JSON export at `/data/atlas-0.1.0.json`.
  • Content and data: all rights reserved (LATEAM AI SAS).