Third-Party Auditing of MindBridge AI Systems

Description

Holistic AI conducted a third-party audit of the technology stack of MindBridge. Their algorithms are vital for internal and external auditors to verify the soundness of financial statements, performance of activities, and many other supervisory needs.

Challenge

To audit MindBridge’s algorithms to Level 7, known as ‘Glass-box’ or ‘White-box’, with full access to the MindBridge algorithms. There is a great deal of uncertainty around the use of Machine Learning in audit. Regulators and audit firms lack the necessary skills to assess the algorithm implementation in MindBridge.

Solution

The audit approach covered three key categories; Privacy, Explainability, and Robustness. Mindbridge's system comprise more than 60 different algorithms contributing to its output. Each of them was audited separately. Correctness relative to documentation, resilience and documentation quality were audited.

Outcome

• A spreadsheet detailing the audit of all the algorithms.
• Individual input files used by Holistic AI to perform resilience and correctness tests for each algorithm.
• Systems were assured and reporting made possible through full documentation.