Real Policy Engine Demos
Interactive Security Scenarios
Live functional demonstrations connected to AgentGuard’s actual policy engine, injection scanner, and audit logger. Every demo executes real backend logic and records auditable forensic traces.
Prompt Security
Scenario A: Adversarial Prompt Injection Defense
The Security Problem
Autonomous agents ingest untrusted user text. Malicious inputs attempt jailbreaks, prompt leaking, or instruction overrides to commandeer the agent.
Expected AgentGuard Behavior
AgentGuard scans the prompt using regex heuristics and token boundaries before downstream actions are taken. Injection payloads are immediately flagged with risk score > 0.8 and BLOCKED.
Configure Scenario Inputs
•
What the Developer Learns
Deterministic pre-execution scanning stops adversarial prompts before tools or payment APIs are ever called. Configured rules target known jailbreaks without relying solely on model self-reflection.
Actual Backend Decision
Click “Run Demo Action” to execute this scenario.
Calls actual AgentGuard policy evaluation API on port 3001