AI auditing
Jakob Mökander · 2025
Artificial intelligence (AI) auditing has recently attracted much attention. Public watchdogs have called for AI systems to be audited; researchers have published guidance on how to audit AI systems, and policymakers have drafted regulations that subject high-risk AI systems to independent audits. But what is AI auditing? Why is it needed? And how can AI systems be audited in practice? This introductory chapter provides an overview of AI auditing as a field of research and practice. In doing so, it offers readers new to the topic three key takeaways. First, AI auditing is a nascent field in which best practices have yet to emerge. As a result, attempts to audit AI systems have much to learn from how audits are structured and conducted in areas like financial accounting and safety engineering. Second, AI audits can – if properly designed and implemented – help identify and mitigate some of the legal, ethical and technical risks AI systems pose. Auditing should thus be viewed as an integral part of holistic and multifaceted policy responses to govern AI. Finally, it is important to remain realistic about what audits can reasonably be expected to achieve. As a governance mechanism, AI auditing is subject to a wide range of theoretical and practical limitations. Understanding and accounting for these limitations is key to designing effective and feasible AI auditing procedures.