Answers to common questions about DKnownAI Guard, agentic AI security, prompt injection detection, and context-aware risk classification.
DKnownAI Guard is a security API built for agentic AI. It helps developers detect manipulation, separate operational and content-related risk, and keep useful agent behavior intact when requests are legitimate.
Traditional content moderation focuses on whether text itself is harmful. DKnownAI Guard focuses first on whether someone is trying to manipulate the agent, then separates operational risk, compliance-sensitive content, and routine safe requests so developers can respond more precisely.
Yes. DKnownAI Guard is designed to detect prompt injection, jailbreak attempts, system prompt extraction, role-based deception, and other tactics used to hijack agent behavior or override intended controls.
AGENT_HACK indicates manipulation targeting the agent. SYS_FLAG indicates a direct request for a high-risk system action. CONTENT_FLAG indicates content that may require compliance review. SAFE indicates a routine request without meaningful risk characteristics.
DKnownAI Guard is designed for AI workflows in which agents execute code, access files, call tools, or interact with internal systems. It fits developer tools, copilots, automation agents, operational assistants, and other workflows where model output can trigger real actions.
Use the same session_id across one continuous conversation when you want DKnownAI Guard to evaluate the latest message together with earlier turns. This is especially useful for detecting multi-turn escalation, role-based social engineering, and context-dependent manipulation.