An attack technique where malicious instructions are embedded in input data to manipulate an AI model's behavior.
Prompt injection is a real security concern for AI applications that process untrusted input — like a document or webpage — since that content could contain hidden instructions the model might follow.
