AWS Certified AI Practitioner AIF-C01 Practice Question

Placing sensitive or proprietary information directly in a prompt can lead to unintended data exposure. Because the model may retain prompts during future training or caching, that information can later resurface in responses to other users. Prompt poisoning refers to inserting misleading content to change a model's behavior, prompt hijacking involves an attacker manipulating an in-flight prompt, and jailbreaking attempts to bypass safety guardrails. None of those describe accidental leakage of private data.