Invisible AI Text Watermark & EU AI Act Compliance Inspector

⚡ Watermark Synthesizer & Pipeline

Synthetic AI Text to Analyze & Watermark 165 chars | 18 tokens

Watermark Scheme

Secret Key Seed (PRNG)

Gamma γ (0.50)

Bias Delta δ (2.0)

Token-Level Green-List Spectrum Inspector Click token to inspect PRNG hash & entropy

Selected Token: #1 "Artificial" GREEN LIST

PRNG Hash (Seed: 42 + Prefix): 0x7a3f89 • Greenlist Partition: Prob 0.50 • Watermark Bias Boost: +2.0 logit

⚔️ Adversarial Perturbation Attacks

Current attack state: Clean Baseline (No adversarial edits applied)

📊 Detection Telemetry & Provenance

WATERMARK CONFIRMED

            Statistical Z-Score
            2.357
            Significance threshold > 1.645 (α=0.05)
          

Hypothesis P-Value 0.0092 Chance of false-positive baseline

Green Tokens 14 / 18 77.8% of evaluated tokens

Tamper Resistance High Survives mild paraphrase

Null Hypothesis (Random) Deterministic Watermark

z=1.65

z = 0.0 z = 2.0 z = 4.0+

EU AI Act Regulation (EU) 2024/1689 • Article 50 Audit

✓
Art. 50(2) Machine-Readable Marking:
Invisible cryptographic watermark is verifiable by automated parsers without human intervention.
✓
Art. 50(2) Tamper-Resistance:
Maintains statistical confidence interval under whitespace/homoglyph perturbations.
✓
Art. 50(4) Provenance Metadata Integrity:
Cryptographic hash anchors generation metadata to generator seed #42.

Decoded Provenance Metadata Payload

eu_provenance://ai_act_art50?gen=craftjs_v4&timestamp=1723685725&sha256=9b7c8e2a

Trace the saved watermark counting rules and fixture limits

Your browser does not support video.

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 inspector offers green-list, zero-width and homoglyph annotation modes. In zero-width mode every even indexed token carries a displayed annotation unless the strip attack is selected. In an illustrative twenty-token sequence, ten annotations become zero after that branch. Both bars show annotated tokens at fifty pixels per token. The written routine assigns classes and flags to token objects; these annotations do not establish a validated steganographic encoding in the raw text. The audit manifest uses a constant fingerprint string with thirty-two hexadecimal characters after its SHA-256 label. A complete two hundred fifty-six bit digest would need sixty-four hexadecimal characters. Both bars count characters at eight pixels per character. The source does not calculate that fingerprint from the input. Preserve its exported value, but do not present its label as an actual digest, legal certificate or cryptographic proof.

⚡ Watermark Synthesizer & Pipeline

📊 Detection Telemetry & Provenance

EU AI Act Article 50 Provenance Audit Certificate

Trace the saved watermark counting rules and fixture limits