The Financial Conduct Authority has published examples of good practice and areas for improvement in financial services AI. Drawing on supervisory work and industry engagement, the publication explains how existing regulatory expectations apply to systems firms develop or buy.
The examples address problems that can be missed in a successful demonstration. They include overstating a system’s capabilities, leaving consumers in inflexible automated journeys and assuming a language model can recognise vulnerability without adequate testing. The FCA also highlights the need for clear accountability and monitoring throughout a system’s life.
For bank and insurance leaders, the practical question is what evidence supports a decision to deploy. A model evaluation is only one part of that answer. Teams also need to understand whether customers can obtain help, who can intervene when a journey goes wrong and how outcomes will be reviewed after launch.
The publication should not be presented as a new standalone AI rulebook. Its importance is the detail it adds to the application of existing obligations. A useful internal response would be to select one live AI journey and compare its governance and customer-support arrangements against the regulator’s examples.
