The correct answer is C.

The continuous evolution of AI models may lead to different behaviors under changing conditions or new data. Unlike static models, AI models can learn and adapt, which means their behavior and performance can shift when they encounter new datasets or economic conditions. This makes ongoing validation a challenge, as a model that may be stable and predictable during one period may behave differently under other conditions, potentially complicating risk assessments and the application of governance principles.

A is incorrect. Increased manual intervention is not inherently problematic, and the challenge with AI models is not tied to their reliability but to their evolving nature.
B is incorrect. Though data volume is substantial for training, the specific challenge pertains to model evolution and validation, not the data quantity itself.
D is incorrect. Establishing accountability is indeed important, but the question focuses on the validation challenges of changing behaviors in AI models, not the operational governance structure.

Things to Remember

• Model Risk Management (MRM) is a crucial aspect in financial institutions to ensure the proper validation and monitoring of all models, including AI models.
• Explainability and interpretability of AI models are essential for risk management, as understanding how the model arrives at its decisions is crucial for validation and risk assessment.
• Data quality and bias in training data can significantly impact the performance and risk associated with AI models, requiring thorough validation processes.
• Ongoing monitoring and backtesting of AI models are necessary to detect any shifts in behavior or performance that may pose risks to the institution.