2. Prompt injection
When malicious instructions are hidden within input (e.g., in a document, website, or email that the AI model processes), causing the model to follow those instructions instead of the original, legitimate task.
Gnostotheca 9325