Gap in AI Audit Standardization

Answer

The main unresolved barrier is the absence of a shared, operationally precise audit regime: there is no agreed set of procedures, taxonomies, standards and auditor qualifications, so audit results cannot be treated as comparable evidence of conformance from one jurisdiction to the next 0fc70194. The corpus shows this barrier appearing at three layers: who may certify an audit 6e40278a, which normative criteria an audit tests 6cf171cd, and how audits are carried out 0fc70194.

Summary

The three verified sources do not contradict one another, but they place the bottleneck at different layers. A 2026 policy comparison finds independent third-party conformity assessment is the dimension where jurisdictions differ most: the EU AI Act (Article 43) mandates external notified-body assessment, whereas Saudi SFDA provisions remain internal quality-management requirements 6e40278a. A 2024 FAccT framework paper identifies missing agreed procedures, taxonomies and quality-assurance guardrails, with practitioners using customized frameworks and ad-hoc tools 0fc70194. A 2021 review states that widely accepted standards have yet to emerge, that there is no consensus on high-level ethical principles, that it is unclear who audits whom, and that the global nature of automated decision systems challenges national jurisdictions 6cf171cd. None of them directly tests mutual recognition between named jurisdictions.

Evidence

• Alqheedan 2026 (One Health) — scored AI-health policy documents from Saudi Arabia, the EU, the USA and China; independent third-party conformity assessment carried the largest weighted gap for Saudi Arabia, and the author reports the coding standard for that dimension had itself been applied inconsistently across jurisdictions. 6e40278a
• Lam et al. 2024 (ACM FAccT) — with supporting work by Ojewale 2025 and Costanza-Chock 2023, finds auditing lacks agreed procedures, taxonomies, standards and quality-assurance guardrails; third-party auditors use customized frameworks. The source states it does not test mutual recognition. 0fc70194
• Mökander et al. 2021 (Science and Engineering Ethics) — review of existing auditing tools concluding that widely accepted standards have yet to emerge; lists lack of institutional clarity about who audits whom and the challenge of national jurisdictions. The paper excludes legal aspects of auditing. 6cf171cd

Methodology

The three facts were read side by side and sorted by the layer of the audit ecosystem at which each locates the barrier (certification authority, normative criteria, operational procedure). They were not pooled into a single ranking because none carries a structured measurement and they use different methods (policy scoring, literature and tooling review, field scan). The 2024 fact is itself a multi-paper summary, so its supporting papers are treated as part of one source.

Limitations

• No source directly tests cross-jurisdiction mutual recognition of audits; the 2024 fact says so explicitly 0fc70194.
• The 2021 review excludes legal conformance, the very setting of the research question 6cf171cd.
• The jurisdiction comparison covers four jurisdictions and the health-AI domain only 6e40278a.
• With three facts and no measurements, the corpus cannot rank the three layers by importance.