Procurement sent an AI vendor due diligence questionnaire: what AI Act evidence answers it
An enterprise buyer's procurement has sent a startup vendor an AI vendor due diligence questionnaire with AI Act items. The buyer is the deployer under Art. 3(4) of Regulation (EU) 2024/1689, and its duties under Art. 26 shape the questions; the vendor is the provider under Art. 3(3) and answers with the instructions for use under Art. 13 and the Annex IV documentation under Art. 11, backed by Art. 15 results, from the Annex III application date Deferred. This page maps the items to articles, gives the sequence and lists the evidence.
Where this starts
Regulation (EU) 2024/1689 reaches the vendor through the buyer's questionnaire before any regulator asks, and no compliance function exists on the vendor's side to answer it: engineering produces the evidence that the buyer's legal reviewer accepts, and one round of testing has to match the items on the questionnaire. On the buyer's side the questionnaire is one output of an inventory: the governance function classifies each bought system, records its role and tier, and wants the vendor's answers in a form that internal audit, risk and the security function read in the same register as the systems built in-house. The startup hub carries the vendor's obligation set with its dates, and the deployer side of the same purchase is on the Art. 26 scenario.
The roles come from Art. 3. The vendor develops the system and places it on the market under its own name, so it is the provider under Art. 3(3); the buyer uses the system under its authority, so it is the deployer under Art. 3(4). Art. 25(1) moves the provider role to the buyer where the buyer puts its own name or trademark on a high-risk system already on the market, makes a substantial modification to it, or changes the intended purpose of a system so that it becomes high-risk; the questionnaire's integration questions settle the role before anything else. The tier comes from Art. 6(2): the system is high-risk where its intended purpose matches an Annex III use case, unless one of the Art. 6(3) conditions applies and the system performs no profiling of natural persons. Where the system is not high-risk, the Art. 26 items fall away and the vendor's answer is the classification record itself; where it is high-risk, the buyer carries Art. 26 and the vendor carries Art. 9 to 17 from the Annex III application date.
Each AI Act item on a questionnaire built from the Act traces to a deployer duty. Art. 26(1) requires the deployer to use the system in accordance with the instructions for use, so the first request is the instructions themselves, which Art. 13(2) requires to accompany the system. Art. 26(2) requires the deployer to assign human oversight to persons with the necessary competence, training and authority, so the questionnaire asks for the oversight measures the provider describes under Art. 13(3)(d). Art. 26(4) makes the deployer responsible for the relevance of input data under its control, so it asks for the input specifications under Art. 13(3)(b)(vi). Art. 26(5) requires monitoring on the basis of the instructions for use, and Art. 26(6) requires the deployer to keep the logs under its control for at least six months, so it asks for the log mechanisms under Art. 13(3)(f). Art. 26(9) lets the deployer use the Art. 13 information for its data protection impact assessment, so the accuracy, robustness and cybersecurity declaration under Art. 13(3)(b)(ii) and the risk circumstances under Art. 13(3)(b)(iii) go into that file. The Act addresses the Annex IV documentation under Art. 11(1) to national competent authorities and notified bodies; a questionnaire that asks for Annex IV extracts is asking for the evidence behind the Art. 13 declarations, and the provider decides what it discloses under its own confidentiality terms.
The accuracy item has no number in the Act. Art. 15(3) requires the provider to declare the level of accuracy and the metrics in the instructions for use, and Art. 15(1) asks for an appropriate level. No harmonised standard is cited in the Official Journal as of the anchor date, so no presumption of conformity is available and the buyer has no cited standard against which to read the figure. The declaration structure in use today is prEN 18229-2Draft, a JTC 21 draft, with prEN 18229-3Draft, a JTC 21 draft with no public text, for the content of the information a provider gives deployers, and NIST AI RMF 1.0Published, a published framework, for the metric selection rationale. The provider sets the target, justifies it against the context of use and records the justification at the target-justification step of PR-15-ROB v1.0; the buyer's reviewer reads that record and the test report together.
The sequence
The work runs in the order below: the roles and the tier first, then the item map, then the instructions for use, then the tests, then the consistency check and the indexed pack.
Record the role of each party under Art. 3 and match the intended purpose against Annex III.
Record Classification and role determination report (PR-CLS output)
[Regulation (EU) 2024/1689, Art. 3(3), 3(4), 25(1) and 6(2); PR-CLS steps 8 to 17]
Map each questionnaire item to the deployer duty behind it and the provider element that answers it.
Record Questionnaire item to article map, one traceability matrix row per item
[Regulation (EU) 2024/1689, Art. 26(1) to 26(9) and Art. 13(3)(a) to (f)]
Mark each Art. 13(3) item present, partial or absent in the instructions for use.
Record Instructions-for-use completeness checklist against Art. 13(3)(a) to (f) (PR-11-DOC output)
[Regulation (EU) 2024/1689, Art. 13(3); PR-11-DOC steps 9 and 12]
Extract the declared levels and record the target justification for each metric before any run.
Record Declared-level extract and target justification record with the source of each target
[Regulation (EU) 2024/1689, Art. 15(3) and Art. 13(3)(b)(ii); PR-15-ROB steps 3 and 8 to 9]
Run the held-out, shifted, perturbation and Art. 15(5) attack sets on the frozen version.
Record Accuracy test report and adversarial and cybersecurity test report (PR-15-ROB output)
[Regulation (EU) 2024/1689, Art. 15(1), 15(4) and 15(5); PR-15-ROB steps 14 to 30]
Compare each declared value in the instructions for use with the test result for the same version.
Record Instructions-for-use to technical documentation consistency report (PR-11-DOC output)
[Regulation (EU) 2024/1689, Art. 13(3)(b)(ii) and (v); PR-11-DOC steps 22 to 26]
Mark each Annex IV row and draft sections 2(g), 3 and 4 from the test reports.
Record Annex IV gap table and section drafts with a source reference on each statement
[Regulation (EU) 2024/1689, Art. 11(1) and Annex IV; PR-11-DOC steps 10 and 14 to 17]
Build the traceability matrix and index the evidence pack to the questionnaire items.
Record Obligation to test to evidence traceability matrix and evidence pack index
[PR-11-DOC step 43; PR-CLS step 44]
Evidence this situation ends with
| Artifact | Answers | Produced by |
|---|---|---|
| Classification and role determination report | Art. 3(3), 3(4), 25(1) and 6(2): the role of each party and the Annex III match, with the Art. 6(3) rationale where claimed | PR-CLS v1.0 |
| Instructions for use (Art. 13(2)) | Art. 13(2) and 13(3)(a) to (f): the document Art. 26(1) requires the deployer to follow | the provider |
| Instructions-for-use completeness checklist against Art. 13(3)(a) to (f) | Art. 13(3): each item present, partial or absent, with its location | PR-11-DOC v1.0 |
| Accuracy test report on held-out and distribution-shifted sets | Art. 15(1) and 15(3): accuracy at or above the declared level and consistent across versions | PR-15-ROB v1.0 |
| Adversarial and cybersecurity test report covering the Art. 15(5) attack list | Art. 15(5): resistance to each named attack class, with the threat model | PR-15-ROB v1.0 |
| Declared metric to test report consistency table | Art. 13(3)(b)(ii) and Annex IV section 2(g): each declared value matches the dated and signed report for the same version | PR-11-DOC v1.0 |
| Instructions-for-use to technical documentation consistency report | Art. 13(3)(b)(i), (b)(vi), (c), (d) and (f): each statement matches its Annex IV section | PR-11-DOC v1.0 |
| Traceability matrix indexed to the questionnaire | One row per questionnaire item: the article paragraph, the test or document, the evidence file and its status | PR-11-DOC v1.0 |
Each artifact follows a specification with a synthetic sample on the deliverables page: the classification report follows the assessment memo specification, the test reports follow the test report specification, the consistency table and the questionnaire rows follow the traceability matrix specification, and the indexed set follows the evidence pack specification on the same page. The traceability matrix tool builds the obligation to test to evidence rows in the browser from the obligation records.
If nothing is done
A high-risk system supplied after the Annex III application date without instructions for use that meet Art. 13(2) and 13(3) fails a Chapter III Section 2 requirement, and the provider obligations in Art. 16(a) and 16(d) cover that requirement and the Annex IV documentation; a market surveillance authority acting under Art. 79(2) requires the provider to take corrective action, to withdraw the system or to recall it within a period the authority sets, and a finding against Art. 16 falls in the penalty tier of Art. 99(4), where the amount in a given case is set by the national authority under the Art. 99(7) criteria. On the buyer's side, a deployer without the instructions for use cannot show use in accordance with them under Art. 26(1) or monitoring on their basis under Art. 26(5), and Art. 26(5) requires the deployer to inform the provider and the market surveillance authority and to suspend use where it has reason to consider that use presents a risk within the meaning of Art. 79(1); a finding against Art. 26 falls in the same Art. 99(4) tier. The provider's action under Art. 20 is corrective action on its own initiative once it has reason to consider that the system does not meet the requirements, and the sequence above produces the information that Art. 21(1) requires the provider to supply on a reasoned request from a competent authority.
Services, guides and tools for this situation
The technical documentation and evidence-pack service runs the completeness and consistency checks and indexes the pack to the questionnaire under PR-11-DOC v1.0, within the limits stated on what we cannot do; the classification and readiness assessment produces the role and tier record under PR-CLS v1.0 where the vendor has none.
- Services
- Guides
- ICP hubs
Questions
What does assessing an AI vendor under the AI Act involve?
A deployer assesses a vendor against the items its own duties under Art. 26 of Regulation (EU) 2024/1689 need: the classification under Art. 6, the instructions for use under Art. 13(2) with the accuracy, robustness and cybersecurity declaration under Art. 13(3)(b)(ii), the oversight measures under Art. 13(3)(d), the input specifications under Art. 13(3)(b)(vi) and the log mechanisms under Art. 13(3)(f). PR-11-DOC steps 22 to 26 check each declared value against a dated test report for the same system version. The compliance decision remains with the deployer and its counsel.
Every inquiry receives a reply within 24 hours. Scope and availability are discussed per request.