A project lead turning a proof of concept into a governed deployment wants to know how prompt changes to a generative AI system should be treated. What is the correct approach?
Select an answer to reveal the explanation.
Short Explanation
If a prompt change can alter how the system behaves, treat it like a material change and reassess it. You can't wait for a new model version to decide something risky happened.
Full Explanation
Treat material prompt changes as a trigger for reassessment, since they can alter the system's behaviour and risk profile as much as a model version change. Prompt changes can materially alter outputs and risk even when the underlying model is unchanged, so they warrant the same reassessment discipline as other material changes. Ignoring prompt changes entirely because only model weights matter is wrong because behaviour-affecting changes are not limited to the model's weights. Reviewing prompt changes only if they increase prompt length is wrong because prompt length has no reliable relationship to the significance of a behavioural change. Treating prompt changes as purely a user-interface concern outside governance scope is wrong because prompts drive system behaviour directly and sit squarely within governance scope. For the exam, remember that generative AI systems can shift materially through prompt changes alone, not only through retraining.