Can AI be Auditable?

Regulatory frameworks like the EU AI Act are establishing auditability as a legal requirement, mandating that AI systems undergo independent assessment for compliance with ethical, legal, and technical standards throughout their lifecycle, signaling a shift toward accountability as a foundational principle rather than an optional practice.

Technical opacity and inconsistent documentation remain major barriers to effective AI auditing, with the paper identifying a lack of standardized audit tools, metrics, and documentation practices as critical obstacles that prevent auditors from meaningfully assessing whether AI systems meet responsible AI standards.

Existing responsible AI frameworks contain conflicting principles that complicate auditability efforts, suggesting that organizations cannot simply follow current guidelines without encountering contradictions that make comprehensive compliance difficult to demonstrate or verify.

Effective AI auditability requires embedding assessment practices into development from the start, rather than treating auditing as a final compliance step—this means developers, regulators, and auditors must collaborate throughout an AI system's lifecycle to ensure ethical and legal alignment.

Building a functional AI audit ecosystem demands international coordination and empowered auditors, as the paper emphasizes that no single organization or country can solve auditability challenges alone, and auditors need sufficient resources, expertise, and independence to conduct meaningful assessments.