Recent finance research points to a practical frontier: AI systems in investment risk and financial crime need provenance, escalation, and audit controls around the model.