Which of the following describes a realistic risk or limitation associated with prompt engineering for generative models?

Prompt engineering involves crafting inputs to guide generative models, but this process can inadvertently create security vulnerabilities. Prompt injection is a significant risk where malicious prompts can manipulate the model's behavior, potentially overriding safety guidelines, extracting sensitive information, or generating harmful content. This makes the model susceptible to unintended actions or misuse. The other options are incorrect because: Prompt engineering does not guarantee consistent, deterministic outputs; generative models are inherently probabilistic. While prompt engineering aims to improve output quality, it does not fully remove risks like data poisoning (which affects training data) or model hijacking (which involves unauthorized control over the model itself, not just its output generation). Prompt engineering does not ensure models will repeatedly produce highly reliable outputs on real-world data; real-world data is diverse and unpredictable, and even well-engineered prompts can struggle with novel or ambiguous inputs.