Glossary · Agentic AI vocabulary
Prompt injection
Content crafted to make a model or agent follow instructions its operator did not intend, including instructions hidden in tickets, documents or device banners an agent reads.
In REAP
In REAP: consequential actions require approval regardless of what the model concludes; retrieved-content handling is described in the security packet.
Related terms
Bring us your last P1.
Then see how REAP connects network evidence to help engineers investigate incidents faster: a 30-minute session, evidence first. The replay on your own network happens inside the proof of value.