LLM01

OWASP Top 10 for LLMs / Prompt Injection

Crafted user input or hidden text alters an LLM's behavior or output in ways the designers did not intend, whether the manipulation is directly typed or buried in content the model reads.

Think of it likeA stranger tucking a fake note into your kid's lunchbox that tells the babysitter to unlock the front door.

In plain English

Attackers hide sneaky instructions in the words an AI reads, tricking it into ignoring its rules and doing things you never asked for.