Why is prompt injection different from ordinary bad input?

Example Answer

The way I'd explain it in an interview is this: Ordinary bad input is mostly a content-quality problem. Prompt injection is a control-plane problem. The attacker is not just providing messy data. They are trying to smuggle instructions through a channel the system should have treated as untrusted content.

That distinction matters because the fix is different. Better prompting may help with ordinary ambiguity, but prompt injection requires separation between instructions, data, tool use, and policy enforcement. If the system cannot tell content from control, it is easy to steer.

So I treat prompt injection less like a rude user and more like an adversarial attempt to reshape how the system reasons and acts.

Common Poor Answer

A weak answer is saying prompt injection is just clever bad input. The important difference is that it attacks system control, not just output quality.