Bailey Warns Banks: Test AI Systems Before and After Deployment

Bank of England Governor Andrew Bailey has called for rigorous testing of artificial intelligence (AI) before and after deployment, emphasizing model assurance and intervention capabilities. In an article published on September 30th, Bailey linked advances in AI to financial stability, including the resilience of banks, payments networks, and financial market infrastructure.

Bailey's argument centers on the need for testing to ensure that AI systems can operate within defined limits and intervene effectively when something goes wrong. He suggests that understanding an AI system's behavior and establishing effective intervention should precede decisions about future regulation architecture.

The Bank of England Governor highlighted the importance of observable behavior in oversight, stating that approval requirements alone offer limited assurance unless tests establish that an AI system pauses at the correct point, presents sufficient information to reviewers, and respects resulting decisions. He also stressed the limits of testing, warning that models will behave unexpectedly and failures cannot be eliminated.