AI models flag falsified audit documentation with 91% accuracy in pilot

Anomaly detection trained on thousands of social audit reports learns to spot coached interviews, cloned time records and template-generated wage data.

A consortium pilot applying machine learning to social audit evidence flagged falsified documentation — duplicated payroll entries, statistically impossible working-hour distributions and copied worker interview responses — with 91% accuracy against known fraud cases.

Audit firms are integrating the tooling into quality assurance, while brands use similar checks on supplier self-assessment questionnaires before accepting declarations at face value.