Back to Markets
Stocks● Neutral
Why Banks Are Rethinking AI Model Governance
Risk managers say AI models force a shift in the three lines of defence, with first-line teams taking on continuous monitoring duties.
Continue with
Why Banks Are Rethinking the Three Lines of Defence for AI Models
The traditional three lines of defence governance model is buckling under the weight of artificial intelligence, according to risk managers at several large banks. The first line – the business units that own the models – needs to take on more formal responsibility for continuous testing and monitoring, they said.
Financial risk models have long relied on a second-line validation team that builds a challenger model and, in some cases, runs part of the testing. That division of labour works when a model is a static code base updated quarterly or annually. AI models are neither static nor easily challenged by a separate team.
Machine learning models retrain themselves on new data. The logic shifts. A validation team that sees the model once a quarter cannot audit a system that has already changed. The first line, which deploys the model day to day, is the only group in a position to catch drift or bias in real time, the risk chiefs argued.
One head of model risk at a European bank said the current framework "was designed for a world where models were built once and validated once. AI inverts that. The validation has to be continuous, and the first line is the only group that can deliver it."