Skip to content

AuditRecord → Governance Evidence Mapping

Informational mapping, last verified 23 August 2026. Not legal advice. ISNAD produces evidence artifacts; it does not certify compliance with any framework. Verify current obligations with qualified counsel.

Using ISNAD does not make you, your product, or your pipeline compliant with the EU AI Act, ISO/IEC 42001, the NIST AI RMF, the SDAIA framework, or any other regulation. This document maps the fields ISNAD emits to the concepts those instruments discuss, so an auditor can find the evidence more easily. It makes no claim of sufficiency.

Reference dates

Instrument Date
EU AI Act — Art. 4 AI literacy in force since February 2025
EU AI Act — Art. 50 transparency duties (deployers) in force since 2 August 2026
Digital Omnibus (dates amended) in force 27 July 2026
EU AI Act — Annex III high-risk obligations 2 December 2027
EU AI Act — Annex I (embedded in regulated products) 2 August 2028

The mapping

Each row maps one AuditRecord field to the governance concepts it can support as evidence. The right-hand columns are not "satisfied by" claims — they are "helps an auditor find" pointers.

AuditRecord field EU AI Act ISO/IEC 42001 (AIMS) NIST AI RMF SDAIA
record_id, generated_at Art. 12 record-keeping (logs) A.7.5 documented information MEASURE (tracking) risk management
claim_id, claim_text Art. 12 traceability; Annex IV tech documentation A.9 system operation MAP (context) data governance
chain[] (who handled the claim, in order) Art. 12 logging; Annex IV supply-chain description A.8.2 AI system lifecycle GOVERN (accountability) model accountability
chain[].upstream_ids[] (the explicit DAG) Annex IV (dependency graph); Art. 12 traceability A.8.2 lifecycle MAP (system structure) model accountability
chain[].grade, grade_rationale Art. 14 human-oversight evidence A.9.3 monitoring MEASURE (test/validation) model accountability
chain[].narrator_type Annex IV (actors); Art. 50 disclosure of AI use A.5 roles & responsibilities GOVERN transparency
chain[].model_identifier, model_version Annex IV (model identity); traceability A.8.1 configuration management MAP (model cards) model accountability
chain[].input_hash, output_hash Art. 12 (immutability of logs) A.7.5 integrity of records MEASURE (reproducibility) data governance
grading_strategy (name, version, parameters) Annex IV (methodology description) A.8.1 configuration management GOVERN (reproducibility) model accountability
weakest_link Annex IV risk description; Art. 9 risk management A.8.4 impact assessment MEASURE (weaknesses) risk management
source_documents[].uri, content_hash, licence Annex IV training-data documentation A.7.2 data quality MAP (data provenance) data governance
human_oversight[] (actor, action, timestamp) Art. 14 human oversight A.5 roles; A.9.3 monitoring GOVERN (oversight) human oversight
environment.isnad_version, python_version, platform Art. 12 (system identity) A.8.1 configuration management GOVERN (reproducibility) transparency
integrity.record_hash Art. 12 (log integrity) A.7.5 integrity GOVERN (auditability) transparency
integrity.detached_signature (reserved) Annex IV (authentication) A.7.5 GOVERN

What ISNAD deliberately does not provide

  • No conformity statement. Nothing in an AuditRecord says "compliant."
  • No risk classification. ISNAD does not decide whether a system is "high-risk" under Annex III — that is the provider's legal determination.
  • No human-oversight decision. ISNAD routes to REVIEW; it does not perform the review.
  • No source-legitimacy attestation. A RELIABLE narrator grade is an operator-assigned judgment, not a verified fact about the upstream source.

Reading the mapping correctly

The columns answer "where does an auditor look for the thing this field evidences?" — not "does this field satisfy that requirement?" A field can be necessary evidence and still be insufficient on its own. If a lawyer, a notified body, or a regulator asks whether ISNAD makes you compliant, the answer is no — and this document exists so that answer is never ambiguous.