White Paper - Developing a Framework for Evaluating LLM Benchmarks

Whitepaper

This is a members only resource.

Please login to access.

Author Wei He PHD
Jason Toh Bcom CA
Jyosmitha Munnangi BTech (Computer Science)
Kathy Condos AMIIA
Date 2026
Topics Explored Internal Audit, Emerging Technologies and Artificial Intellegence
Extract/Description Internal audit teams are increasingly experimenting with and adopting Large Language Models (LLMs) to perform internal audit and assurance activities. However, there is little clarity on how internal auditors can determine which LLMs should be selected for these activities.

This paper proposes a capability-first framework that maps core internal audit activities, grounded in the ‘Global Internal Audit Standards’ (GIAS), against existing, externally available LLM benchmarks and internal audit lifecycle tasks.
Key Points
  1. A design for the framework is outlined including key inputs, benchmarks and weighting assumptions.
  2. Internal audit capabilities included in the framework are explored and mapped to benchmarks.
  3. A case study showcases the use of the framework and key findings for internal audit functions to learn from.
Relevant Industries All
Level of Assumed Knowledge Expert
Aligned to Global Internal Audit Standards Yes