問題文
An insurer already has a mature model risk management program for its actuarial statistical models. It now plans to deploy a large language model that drafts claim denial letters. Which difference in risk profile should the risk practitioner highlight to the model risk committee?
選択肢
- The language model requires more historical data than the actuarial models so the main new risk is data volume.
- The language model is deterministic so once it passes validation the same input will always produce the same letter and no ongoing review is needed. This makes its risk profile lower than that of the actuarial models.
- The language model has no regulatory exposure because it only drafts letters and a human signs them so the actuarial governance process can be reused unchanged.
- The language model produces free-form text whose failure modes are open-ended so output-level review and content guardrails matter more than the input-coefficient checks used for the actuarial models.