What does prompt injection involve in AI security?

• A. Prompt injection involves evaluating and sanitizing data input to prevent abuse of AI models.
• B. Prompt injection is the process of increasing model capacity.
• C. Prompt injection is about storing prompts securely.
• D. Prompt injection refers to injecting prompts into hardware firmware.

Answer: (A)

Prompt injection is about stopping adversarial prompts from steering an AI model into unsafe or unintended behavior. The best approach to this risk is to evaluate and sanitize input data so dangerous instructions or hidden prompts cannot influence the model. By validating prompts, removing or neutralizing harmful tokens, enforcing strict input rules, and applying guardrails before the model processes anything, you reduce the chance that a crafted prompt will cause abuse or leakage of restricted information. This defensive stance—cleaning and controlling what the model sees as input—is central to mitigating prompt-based attacks. The other ideas, like increasing model capacity, storing prompts securely, or injecting prompts into hardware firmware, don’t address the manipulation of user-supplied prompts at the input stage.