Generative AI (Gen AI) is transforming model validation in the banking sector

Banks are increasingly turning to generative artificial intelligence (Gen AI) to speed up processes, support decision-making and process complex information. This gives rise to new risks that are pushing established validation processes to their limits. Our English-language white paper, ‘Validation of Generative AI Models in the Banking Sector’, shows how banks should further develop their model risk management and which additional testing and governance mechanisms will be necessary to use Gen AI in a safer, more controlled and more regulatorily robust manner.

Gen AI not only delivers numeric predictions but also generates text, recommendations and complex inferences. As a result, risk profiles are shifting and changing. Banks should understand what new risks Gen AI creates, how existing risks are changing, and what adequate control and monitoring mechanisms are now necessary.

Gen AI models lack transparency and can generate biased, incomplete or simply incorrect content. This poses operational, regulatory and reputational risks, for example. Unlike traditional models, Gen AI systems are less deterministic, making quality control more challenging. Validation must systematically account for the likelihood of hallucinations, bias and erroneous conclusions – regardless of how convincing the results may appear.

Legislators and regulatory authorities are raising the bar in terms of transparency and oversight. Regulations such as the EU AI Act, the AI Framework from the US National Institute of Standards and Technology (NIST) or the US Federal Reserve’s SR 11-7 supervisory guideline require traceable decisions, clear responsibilities and continuous monitoring – particularly in high-risk use cases such as lending, compliance analyses or risk assessments. For banks, this means that Gen AI should not only be powerful, but also documented, explainable and verifiable by third parties.