System Prompt
(Benign + Hidden + Oversight)
Baseline Output
(Honest Reference)
Agent Interaction
(VFS Sandbox)
Reasoning Trace
Model Output
Prong 1: Regex
(Fact Verification)
Prong 2: Blackbox
(Grok-4.1 Judge)
Prong 3: Glassbox
(GPT-4.1 Judge)
Baseline Reference
Deception
Profile