← Back to Demos Lab

Prompt Injection & Jailbreak Gate

Reproduction of realZachi/typesafe-guard. Tests adversarial prompt inputs against injection attacks, system prompt theft, and destructive commands in 22 ms before hitting your LLM.

SAFE (ALLOW) Safety Decision

22 ms Gate Latency

0 tokens Token Overhead

$0.00004 Cost per Evaluation

1. Test Prompt Input

Select an adversarial sample or enter custom untrusted user input:

2. Calibrated Decision Tensor

Probabilistic risk scores evaluated via Jev noul and score primitives:

Direct Prompt Injection

0.02

System Prompt Extraction

0.01

Destructive Action Risk

0.01

Refusal Recommendation

0.03

✓ Input Verified Safe

No adversarial instructions or destructive commands detected. Evaluated safely in 22 ms. Proceeding to frontier reasoning model.