Indirect Prompt Injection: How to Protect an AI Assistant from Instructions Hidden in Documents
Practical layers of defence against instructions planted in the files and pages an assistant reads: permissions before retrieval, separating untrusted content, restricting tools, human approval and testing.