What is Prompt injection?

Prompt injection is when someone sneaks harmful or misleading instructions into text that an AI will read, so the AI might ignore the original rules and do something it should not. This can happen through a webpage, a document, or a message that the AI is asked to summarize or follow. For everyday users, it matters because an AI tool could give unsafe advice, reveal private info it was shown, or behave in unexpected ways. A downside is it can be used to trick AI systems, so trustworthy tools add safeguards and users should be careful about what content they feed into AI.

Sometimes a computer helper reads instructions that someone else wrote, not just yours. A bad person might hide a bad instruction inside a message to confuse the helper. Then the helper might follow the hidden instruction instead of doing what you asked.

Imagine you ask a friend to follow a recipe you wrote, but someone secretly slips an extra note into the paper that says, “Ignore the recipe and add way too much salt.” If your friend does not notice the sneaky note, the meal turns out bad. Prompt injection is like that hidden note, but for AI instructions.