Glossary: Application and Credential Attacks, and AI Risks

indirect prompt injection

Prompt injection hidden in content the AI reads for someone else, such as white text in an email to Priya telling her assistant to forward invoices. The honest user never sees the instructions.