What supervision must be able to see
Monitoring only the model would miss much of the risk. Supervisors need visibility into training and customer data, third-party dependencies, decision thresholds, overrides, incident logs, post-deployment changes, and the business processes wrapped around an AI system. A reliable tool can still produce harmful outcomes when incentives, data, or operating controls are wrong.
The standard should be evidence rather than an AI label. Banks and insurers should be able to show which decisions are assisted or automated, how customers can challenge an outcome, who owns the risk, what happens during outages or drift, and whether a vendor can be replaced without disrupting essential financial services.
Go to the source
Read the evidence behind this analysis. External links open in a new tab.
Reuters — Germany’s watchdog to monitor AI at banks and insurers
