Glossary · Security

Prompt Injection Defense

Techniques for preventing untrusted input (a document, webpage, or tool result) from being interpreted as instructions by an agent, such as input sanitization, privilege separation, and output filtering.