Fraud & attack lab
Everything below uses your text and the real NexusGuard evaluate API — same rules as Security center. Results are logged to the audit trail.
Your scenario
Natural language transfer, bill pay, or paper-trade wording.
Quick fills (edit after load)
About prompt injection
The rule-based parser does not “execute” hidden instructions — it extracts transfer intent only. For a malicious narrative, use Quick fill → Prompt-injection style and compare the parsed intent to the story you typed. Tightening this further is what full LLM guardrails + policy are for.