An attack technique that embeds malicious instructions in input to make an AI behave in unintended ways.
An attack that mixes crafted instructions into a prompt to make an AI model produce output the developer didn't intend, leak sensitive information, or perform unauthorized actions. A vulnerability specific to LLM-powered applications, ranked at the top of OWASP's Top 10 for LLM Applications.
© 2026 ITBGM