EU AI Act
Assessor Workpaper

Instructions

This workpaper provides structured test procedures for verifying SWT3 witness anchor evidence during a conformity assessment under Regulation (EU) 2024/1689 (the EU Artificial Intelligence Act). Each entry maps an article obligation to the SWT3 procedures that produce evidence, defines what to test, what to expect, and how to determine pass/fail.

How to verify an anchor: Paste the anchor string into the public verifier at sovereign.tenova.io/verify (no account required), or recompute the fingerprint yourself using SHA-256. See Verify an Anchor in 5 Seconds for the full procedure.

Coverage badges: Full = SWT3 procedures directly satisfy the article requirement with cryptographic evidence. Partial = procedures cover the operational/runtime aspect of the requirement; supplementary evidence (policy documents, risk narratives, organizational governance) must be verified separately by the assessor.

Sampling guidance: For sample size, red flags, and CSV query instructions, see Assessor Sampling Instructions.

Factor conventions: Factor legends in this workpaper (e.g., "factor_a = guardrails required") reflect the SWT3 SDK default conventions. Clients may customize factor usage. Before relying on factor interpretations, confirm with the provider that their implementation follows these defaults, or request their factor mapping document.

Methodology boundary: This workpaper documents evidence format and suggests evaluation criteria. It does not prescribe assessment methodology. Conformity determinations remain the assessor's professional responsibility under their applicable accreditation requirements and the Notified Body's designation scope.

Not applicable entries: If a workpaper entry is not applicable to the assessment scope (e.g., the system is not classified as high-risk, so WP-01 through WP-07 do not apply), mark the result as N/A with justification in the Notes field.

Supplementary evidence: Each workpaper entry includes a blue "Supplementary Evidence Required" box listing what SWT3 does NOT cover for that article. These items require separate assessment through document review, interview, or inspection.

Estimated duration: Allow 2-3 hours for a high-risk track review (WP-01 through WP-07) including verification of 5-10 sample anchors. GPAI evaluation (WP-08, WP-09) adds 1 hour. Provider/deployer obligations (WP-10 through WP-14) add 1-2 hours. For initial conformity assessments, plan an additional 1-2 hours for provider orientation and factor mapping confirmation.

Anchor anatomy: Every SWT3 Witness Anchor follows a deterministic format. Here is an annotated example for an EU AI Act-relevant procedure:

Articles 9-15: High-Risk AI System Requirements

These requirements apply to high-risk AI systems classified under Article 6 (Annex III standalone or Annex I safety components). Enforcement: 2 December 2027 (Annex III) / 2 August 2028 (Annex I).

Articles 53, 55: General-Purpose AI Model Obligations

These obligations apply to providers of general-purpose AI models (Art. 3(63)). Enforceable Now -- since 2 August 2025.

Articles 16, 17, 26, 27, 50: Provider and Deployer Obligations

These articles establish organizational obligations for providers placing high-risk AI systems on the market and deployers putting them into use.

Supplementary Evidence Checklist

SWT3 provides cryptographic evidence of operational facts. The following table identifies what SWT3 does NOT cover for each article -- items the assessor must verify through document review, interview, or inspection separately from anchor evidence.

How to Cite Verified Evidence

When referencing SWT3 anchor evidence in your conformity assessment report or audit findings, use a consistent citation format. The following templates cover individual anchor verification and workpaper-level summaries.

Common Assessment Questions

Questions that arise during EU AI Act conformity assessments, with guidance on how to proceed.

What if the system is classified as both high-risk AND uses a GPAI model?
Both tracks apply. The GPAI model provider must comply with Art. 53 (and Art. 55 if systemic risk), while the high-risk AI system provider or deployer must comply with Art. 9-15. These may be different entities. Use WP-01 through WP-07 for the high-risk system assessment and WP-08 through WP-09 for the GPAI model assessment. If the same entity is both GPAI provider and high-risk system provider, all workpaper entries are relevant.

What if harmonised standards are not yet available?
Until harmonised European standards (hENs) are formally designated in the Official Journal, there is no presumption of conformity. Use current SWT3 procedure mappings as interim evidence of operational compliance against the article requirements directly. Update the assessment approach when hENs are published and formally designated.

What about the Omnibus deferral dates?
The Digital Omnibus Regulation deferred enforcement of high-risk AI requirements. Annex III (standalone high-risk): 2 December 2027. Annex I (safety component high-risk): 2 August 2028. GPAI obligations (Art. 53, 55) are already enforceable since 2 August 2025 and are NOT deferred.

What if the provider and deployer are the same entity?
Both Art. 16 (provider obligations) and Art. 26 (deployer obligations) apply. All workpaper entries WP-10 through WP-14 are relevant. The entity must demonstrate compliance with both sets of obligations, though evidence may overlap.

What if the GPAI model is open-weight?
The Art. 53(2) exemption applies to certain obligations for open-weight GPAI models, but it does NOT exempt providers from Art. 53(1)(a) (technical documentation) or Art. 53(1)(c) (copyright policy). If the model is classified with systemic risk under Art. 51, the exemption does not apply at all -- all Art. 55 obligations remain in full force.

What if an anchor fails fingerprint verification?
A verification failure means the anchor has been modified since minting, or the verification inputs are incorrect. First, confirm you have the correct inputs (tenant ID, procedure ID, factors, millisecond timestamp). If inputs are correct and verification still fails, this is a significant finding -- document it and request the provider's explanation. A failed fingerprint may indicate evidence tampering.

Summary Sheet

Record the assessment result for each workpaper entry. Use this as the summary page for the conformity assessment file.