Indirect prompt injection
An attack in which malicious instructions are hidden in content the model reads (a web page, email, document or tool result) rather than typed by the user.
Indirect injection is the main risk for retrieval and agent systems because the model processes attacker-controlled text as part of its normal work. Benchmarks such as OPI, ASB and StruQ measure how often hidden instructions succeed; IBM's LLMON previews reported near-zero success on OPI in November 2026, while cautioning that the resistance is learned rather than enforced and that adaptive multi-turn attacks remain untested.
Related terms
AgentAgent harnessGuardrailsNon-human identityPrompt injection
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.