Invisible AI Text Watermarking & Detection Workbench

EU AI Act Art. 50

Statistical token vocabulary shifting (Kirchenbauer algorithm), real-time logit inspection, and z-score verification.

1. Token Watermark Generator

Green List Ratio (γ): 0.50

Watermark Bias (δ): 2.00

Random Seed Key:

Generated Synthetic Text (Token Breakdown)

• Green Token • Red Token

Logit Probability Shift Visualizer Token: 'process'

Kirchenbauer partitioning uses key hash of previous token to split vocabulary into Green List (promoted by +δ) and Red List.

Green Token Ratio

Green Count / Total

0 / 0

Avg Delta Shift

+0.0

2. Robustness & Human Mutation Tester

Simulate human editing, synonym replacement, or text paraphrasing to evaluate statistical detection persistence.

3. Statistical Z-Score Detection Engine

Calculated Z-Score

0.00

p-value Significance

1.0000

Threshold (α = 0.01)

Z ≥ 2.33

Z-Score Equation: Z = (S_g - γ·N) / &sqrt;(γ·(1 - γ)·N)