standards.sgit.ai / eu-ai-act / provisions / art-009-para-05

Article 9(5) — judged to be acceptable Law

A residual-risk provision that imposes a judgement and does not supply the standard for making it — which is the most interesting kind of obligation there is, and the hardest to evidence.

The operative text is not reproduced on this page. The reading below is derived from secondary sources, which the project flagged against itself before this site existed. A page whose thesis is point at the provision does not get to quote a provision it has not re-derived — so this page publishes the citation, the address, the hashes and the reading, marked as a reading, and points at the authority for the text. The re-derivation status →
Regulation (EU) 2024/1689 · Article 9 · paragraph 5

Article 9(5) — judged to be acceptable

Operative text not reproduced here — see the note above. Read it in the vault, where it was parsed from Formex XML and hash-verified, or at the authority.

pos 577915560498b17d sha not computed — text not re-derived alias art-009-para-05 retrieval: EUR-Lex ELI, via CELLAR Formex XML into the vault

The reading

The obligation to make the judgement is imposed; the standard against which to judge is not supplied.

This is a common and deliberate legislative move, and it is where a graph earns its keep. A provision that says residual risk must be judged acceptable creates a duty whose discharge cannot be checked against the instrument alone — you have to go outside it, to a framework, to a state of the art, to an organisation's own stated appetite. That is a bridge, and it is exactly the kind of edge crosswalks exist to carry.

It is also the point where this site hands over to risks.sgit.ai. The distinction that site draws — accepted is not the same as acceptable — is the whole content of the word this paragraph turns on. A risk somebody signed off is accepted. Whether it is acceptable is a different question with a different answer, and conflating the two is how an acceptance record becomes a compliance claim.

The self-correction that produced a site-wide rule

An earlier reading of this paragraph described it as referring back to paragraph 2. The operative text refers back to “paragraph 2, point (d)”.

One point, not a whole paragraph. The scope is materially narrower, and the error is invisible in prose and permanent in a graph. It produced the rule that now applies to every citation on this site: every provision reference carries its instrument, article, paragraph and point. “Article 9(5)” is not a citation. The rule →

Recorded here rather than quietly fixed, because the correction is more instructive than the corrected text: it is the exact failure mode this site's method is designed to catch, caught by the method, in its own material.

What would ground a claim under this paragraph

RungWhat it would have to beState
FindingResidual risk was judged, by a named party, against a stated standardUnanswered in every example on this site
FactA judgement was recorded, with its date and its basisUnanswered
EvidenceThe record itself — a document, a decision, a sign-offUnanswered
MeasureAn observation of the system's actual residual behaviour, not of the paperwork about itUnanswered

Four ghosted rows is not a failure of this page. An unanswered question is an output — and a provision that imposes a judgement without supplying its standard is one that will show four ghosted rows for most organisations most of the time. That is the finding.

Crosswalk edges

None

No bridge attaches to this provision, in either direction. Why there are none, and what has to be true before there are →

Citing this provision

https://standards.sgit.ai/eu-ai-act/provisions/art-009-para-05.html
positional hash: 577915560498b17d   (sha256 of "eu-ai-act/art-009-para-05", first 16 hex)

The positional hash is the canonical identity; the alias is a projection of it. If the Act renumbers, the alias moves and the hash does not — which is what lets a crosswalk drawn today survive a consolidation. The scheme →