Cryptographic AI Security

AI Text Watermark & Entropy Inspector

★ EU AI Act Synthetic Content Audit Engine (Art. 50 Compliant)
WATERMARK DETECTED
z-Score Confidence 3.12
p-Value (p) 0.0009
Green Ratio (|S_g| / N) 14 / 22 (63.6%)
Expected Green (μ) 5.50
Token-Level Green/Red List Partition Heatmap 22 Evaluation Tokens

Hover over or click individual tokens to view their cryptographic pseudo-random green/red assignment based on the hash of preceding token n-grams.

Green List Token (+δ Boosted)
Red List Token (Unmodified Probability)
Context Anchor (N-Gram Seed)

Theoretical Null Distribution H₀ (Unwatermarked Human Text) vs. Observed Watermarked Sample Indicator.

Receiver Operating Characteristic (ROC) trade-off across varying logit bias δ levels and edit distance attack ratios.

Degradation Telemetry
Current Total Tokens (N): 22
Observed Green Count (|S_g|): 14
Observed Red Count (|S_r|): 8
Current Calculated z-Score: 3.12
Statistical p-Value: 0.0009
Tamper Robustness Margin: HIGH (+1.48 z)
Compliance Audit Evidence Export

Trace the saved watermark counting rules and fixture limits

Read the explanation

The written detector compares green count to token count times the assigned green fraction, then divides by square root of count times fraction times one minus fraction. With an illustrative one hundred evaluated tokens and fraction one quarter, expected count is twenty-five and standard deviation about four point three three. Thirty-five green tokens give z about two point three zero nine; forty-five give four point six one nine. Both bars depict z on one hundred thirty pixels per unit. These illustrative counts explain the algebra, not a measured false-positive rate or proof of authorship. The text entropy source excludes its first token as context before counting evaluation tokens. A ten-token passage therefore has nine evaluated tokens; a twenty-token passage has nineteen. Both bars measure evaluated counts at thirty pixels per token. Classification hashes the preceding cleaned word and the chosen key, then compares to gamma. Delta does not enter that ordinary membership calculation. Changing a displayed bias control is not equivalent to generating new text with a language model. The default transparency passage has an explicit source override when gamma is point two five, delta two and key forty-two. It assigns z three point one two, p point zero zero zero nine, fourteen green and eight red. These bars show the assigned counts on one forty-pixel scale. This is a preset branch, not a fresh statistical measurement. Outside that condition it uses the general written formula. Preserve the distinction between a pedagogical fixture, claimed compliance labels and an independently established detection result.

Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.