indirect prompt injection
/ˌɪndəˈrekt ˈprɒmpt ɪnˌdʒekʃən/
adversarial instructions hidden in external content that a model reads through browsing, retrieval, files, or tools
indirect prompt injection in a sentence
“A webpage contained an indirect prompt injection telling the research agent to upload its notes.”
Origin of indirect prompt injection
Latin indirectus not straight + promptus ready + inicere to throw into
Related Words
prompt leakage
the unintended disclosure of hidden instructions, private context, or other prompt contents
untrusted content
data from a source that is not authorized to issue instructions and may be inaccurate or malicious
guardrail
a policy or technical control that limits unsafe, unauthorized, or out-of-scope behavior
security sandbox
an isolated environment that restricts a program's access and limits the impact of mistakes
least privilege
the security principle of granting only the permissions needed for a specific task
autonomy calibration
matching an AI system's freedom to act with the task's risk, reversibility, and uncertainty