When Auditors Come for Your AI Algorithms
InformationWeek, Tuesday, October 27th, 2026
Auditors are coming for AI. Gartner, PwC and KPMG explain how to build records, controls and accountability that make AI decisions defensible.
Audits depend on predictable processes, but LLMs can answer the same question differently, and prompt logs do not capture the reasoning in between.
PwC's Jennifer Kosar advises recording the process as the system runs and using a three-layer model of design-time risk assessment, periodic review and independent assessment.
KPMG's Mike Sawyer says auditors will ask what data was used, how outputs were evaluated, who reviewed them and how they shaped the decision, while a Gartner survey of 697 AI leaders found two-thirds rate bias monitoring and output accuracy difficult.
Experts urge treating audit readiness as a business capability and acting now, as embedded AI features can quietly break existing controls.