standards.sgit.ai / eu-ai-act / status

Status: what is done, what is derived, what is stale

Every instrument folder on this site has one of these and it is not optional. This one exists because the AI Act material carries two real weaknesses that the project flagged against itself before this site was commissioned, and a page that omitted them would be a worse page.

Weakness 1 — the citations are secondarily sourced

The question set is derived from the structure and subject matter of the named articles as described in secondary sources rather than from a clause-by-clause reading of the operative text.

And, on itself, in the same corpus:

This brief describes the Article 10 change from secondary analysis, which is exactly the second-hand pattern it criticises elsewhere.
Why this matters more here than it would anywhere else. The thesis of this site is that a claim should point at a provision instead of asserting one. A site that ships provisions sourced from commentary has not done that — it has moved the assertion one step away and hidden it behind a citation, which is worse than asserting openly.

The evidence that this is a live hazard rather than a theoretical one is the project's own staleness probe, which found three states of AI Act text circulating in the wild — current, superseded, and a text that never was the law, a negotiating draft published as though in force. And the corollary that removes the obvious mitigation: “Cross-checking two sources cannot establish currency, since agreement only shows both are old.”

WhatState
Graph structure — articles, paragraphs, points, recitals, annexesParsed from official Formex XML retrieved from CELLAR, hash-verified, parsed deterministically. This layer is sound
SHA-256 provenance on retrieved bytesPresent on every node in the vault
Article-level readings and the question setSecondary. Derived from analysis of the articles, not from the operative text
Operative text on this site's provision pagesNot reproduced — deliberately, until re-derived. Each page says so
The Article 99 third penalty tierDisputed internally (1% vs 1.5%) and not published until checked

What fixes it: a clause-by-clause pass over the operative text for every article this site cites, with the retrieval recorded in the vault's source/RETRIEVAL.md, and the content hash computed per provision. Until then, every reading here is marked as a reading and no operative text is reproduced.

Weakness 2 — two commits, and no amendment history

The single most valuable property of this instrument is the one least demonstrated. The business argument for a standards corpus is that the amendment is the business model — the value is in tracking change over time. The Regulation Graph vault has two commits, and it composes Regulation (EU) 2026/1744 in rather than tracking it as a change.

So the artefact that exists is a very good snapshot of a composed text. What it is not, yet, is a record of how that text got there — which is the part that would be worth paying for.

What fixes it: the amendment differ (T3) — two versions in, changed provisions out, computed by content hash so the diff is exact rather than textual. It needs two versions in the vault to compare, which means the pre-amendment text has to be retrieved and stored as its own version rather than superseded. That is the missing commit.

The labelling obligation on an unofficial consolidation

Composing 2024/1689 with 2026/1744 fills a real gap: no official consolidated version exists, and one is genuinely useful. It also produces an unofficial text that looks official, which is the precise shape of the hazard the staleness probe found in the wild.

What this site does about it today: the composition is labelled as a composition everywhere it appears, and the amendment is a first-class node rather than silently applied. What is not settled: what the labelling obligation formally requires — how prominent, in what form, and whether a machine-readable flag is owed as well as a human-readable one. Q5, open →

The conformity gap

None of the deliverables currently grants presumption of conformity, because none has been cited in the Official Journal. Both conditions are required and neither is met.

Worth stating on a status page because it is the thing most likely to be assumed the other way. Conformity with a harmonised standard grants a presumption of conformity with the Regulation — once such a standard exists and has been cited in the OJ. Neither has happened. Any plan that routes compliance through a harmonised standard is routing it through a mechanism that is not available yet.

Summary

LayerState
Structure and provenance of the retrieved bytessound
Reading layer — question set, penalty analysis, worked examplepublished, marked as secondary
Operative text on provision pageswithheld pending re-derivation
Amendment-over-time historyabsent — needs a second version in the vault
Crosswalks out of this instrumentnone exist
Zip and SQLite distributionnot built, on any vault