Security Lab

The Security Lab is an interactive console for testing how the scoring engine responds to a prompt. You type a prompt, the appliance scores it exactly as it would a live intercepted request, and you see the full verdict plus, optionally, the replies real AI agents would give. It is the best place to understand your policies and to sanity-check the engine.

What you see

An input where you enter a prompt and a result area showing the verdict. A verdict has several parts:

• Classification: benign, suspicious or malicious.
• Combined action: SAFE, MONITOR or BLOCK.
• Composite score: a 0 to 100 risk number.
• Total violations and max severity.
• Detection chips showing which detectors fired: FPGA, SUBSTR, KW, FUZZY, DEOBF, PII (see the Glossary).

Two worked examples illustrate the range:

• A benign prompt such as "capital of Texas" returns benign / SAFE, composite 0.00.
• A jailbreak-style prompt asking for instructions to build a weapon returns malicious / BLOCK, composite 70.00, with FPGA, SUBSTR, FUZZY and DEOBF firing.

Where provider keys are configured (Settings), the Lab can also show the actual replies from agents such as Grok and Gemini alongside the verdict.

How to use it

1. Type or paste a prompt to test.
2. Run it and read the verdict: classification, action, score, and which detectors fired.
3. Use it to validate a new or edited policy: craft a prompt that should trip it and confirm the expected detector appears.
4. Compare agent replies where keys are set.

Running a test does not change configuration, so the Lab is available to both roles. Editing the policies that drive the verdicts is Administrator-only.