An ecommerce site will publish an AI-powered chatbot that accepts customer input around the clock. Which input-side vulnerability must be addressed before launch?

Prompt injection is a critical input-side vulnerability where malicious users manipulate the chatbot's behavior by crafting specific inputs that override or bypass its intended instructions. This could lead to the chatbot revealing sensitive information, generating inappropriate content, or performing unintended actions. Data leakage is an output-side vulnerability, where the model inadvertently exposes private training data. LLM hallucinations are an inherent model behavior where the model generates factually incorrect but plausible-sounding information, not an input vulnerability. Concept drift refers to changes in the statistical properties of the target variable over time, affecting model accuracy, and is not an input-side vulnerability.