II.
JailbreakPattern overview
Reference · livejailbreak:prompt-injection-indirect
Indirect prompt injection via retrieved content overview
Adversarial instructions hidden in retrieved web pages, PDFs, or emails that the agent ingests as context.
Attributes
displayName
Indirect prompt injection via retrieved content
technique
indirect-injection
severity
critical
signature
Retrieved documents containing imperative instructions targeting the
assistant ("Ignore previous instructions and ...").
description
Adversarial instructions hidden in retrieved web pages, PDFs, or emails
that the agent ingests as context.
Outgoing edges
None.
Incoming edges
detects_jailbreak1
- content-policy:default-acceptable-use·ContentPolicyDefault acceptable-use policy