Regulators are weighing mandatory independent audits for AI systems used in high-stakes decisions such as lending, hiring and insurance underwriting, according to a discussion paper released this week.
The proposal would require firms to document model training data, testing methodology and known failure modes, with audits conducted by accredited third parties.
“It's the same logic as financial audits — if a model is making consequential decisions, someone independent should be able to check the working,” one policy adviser said.