← InsightBridge Global Press
AUTHOR SERVICES · PATENT PENDING

Dynamic Semantic Fingerprint™

One platform, two protections: every copy is encrypted — and every copy can be traced.

A protection and tracing service for valuable documents. Upload a manuscript; we return an edition that is encrypted against tampering and carries an invisible fingerprint unique to each buyer. Any leak — a copied paragraph, an AI-rewritten excerpt, even a photo of a printed page — can be traced back to the exact person who leaked it. Built for independent authors and for institutions alike: publishers, universities, research institutes, law firms, and any organization whose documents carry real intellectual property.

THE PROBLEM · 为什么需要它

Piracy today is a copy-paste — and every traditional protection dies at the clipboard

The main channel of book piracy is no longer file sharing: it is copying text straight off the screen — from a web page, an e-reader, a phone — and pasting it elsewhere. Visible watermarks are cropped away in seconds. File-level "invisible" watermarks live inside the e-book file itself, so the moment text is copied out, reformatted, OCR-scanned or paraphrased by AI, they are gone forever. Hard DRM locks honest readers into clunky apps and is routinely cracked. When a leaked excerpt surfaces online, the author usually cannot prove whose copy it came from. One untraceable leak can erase years of work.

WHY IT IS DIFFERENT · 五大能力

Five capabilities no traditional product offers together

NEW · BENCHMARKED AUG 2026

It survives AI paraphrasing

The #1 piracy channel of the AI era — large-model rewrite attacks

The most dangerous piracy method today is not copying — it is feeding your book into a large AI model and having it rewritten. We benchmarked exactly that: 6 frontier models (Claude, GPT-4o, Gemini, DeepSeek, Qwen, Kimi) × 4 attack levels (verbatim excerpt, single-pass rewrite, two-model relay, ten-pass rewrite) × 3 text lengths — 5,940 samples in total. Result: AUC 0.92–1.00, and on held-out calibration the system flags 90.7–100% of AI-rewritten leaks at a 1% false-positive rate. Even after ten successive rewrites relayed across two different models, the fingerprint still identifies the source buyer. Traditional watermarks are destroyed by a single AI rewrite pass — ours is not.

6
frontier AI models tested
5,940
attack samples
AUC 0.92–1.00
across all models & attack levels
1%
false-positive rate
I

It survives the clipboard

Copy-paste is still traceable

Today’s main piracy channel is copying text straight off a web page or reader and pasting it elsewhere. Traditional watermarks die at the clipboard. Ours lives in the language itself, so a copied-and-pasted excerpt — from desktop or mobile — still identifies the source buyer.

NEW
II

It survives paper

Print it, scan it, photograph it — still traceable

Our latest milestone: even when a document is printed on a laser printer, then scanned on a flatbed scanner or photographed with a phone, the fingerprint is still recovered and the leaker identified. In production-grade tests, a 14-page scan and desk photos yielded exact buyer identification, and a mixed-evidence examination traced 606/606 lines of a full book. No competitor’s watermark survives the print-scan-photo channel.

III

It collects nothing

Zero personal data — burn after reading

No personal information is required: upload your document, download the protected edition, done. Your original manuscript is destroyed immediately and irreversibly after rendering — only an irreversible fingerprint (no original text) is kept for future tracing. Nothing to breach, nothing to leak.

OUR PLEDGE
IV

No Claude models. Ever.

Your words stay yours — free of any third-party AI watermark

We refuse to use any Claude (Anthropic) model anywhere in our pipeline — none, period. Since August 2026, Claude models embed watermarks into the text they touch; a book processed through them could later be flagged as "AI-generated", putting the author in the impossible position of proving they wrote their own work. That risk is unacceptable to us. Our systems contain no Claude models and have no relationship with them: the ONLY mark ever added to your book is the fingerprint you commissioned — invisible, under your sole control, and removable by no one but you.

WHAT IT DOES · 双重保护

Two protections in one pipeline

01

Document Protection

AES-256-GCM encrypted container

Every fingerprinted edition can be delivered as a sealed, tamper-evident encrypted container (chunked AES-256-GCM with order-bound authentication + AES-KW key wrapping). Any tampering, block reordering, deletion or truncation is detected — not silently passed. Your original manuscript is destroyed after rendering ("burn after reading"), and every render is sealed into a Merkle evidence chain suitable for court exhibits.

  • Tamper, reordering & truncation all detected (4/4 attack classes caught in testing)
  • Original manuscript destroyed after rendering
  • Merkle-sealed forensic evidence chain
02

Buyer Tracing

Invisible semantic fingerprint

Without changing a single word of the reading experience, a buyer-unique signature is woven into the probability of synonym choices and sentence-structure topology. Because the fingerprint lives in the language itself — not in the file — it survives copy-paste, reformatting, format conversion and OCR re-typing. Paste any suspect excerpt into the console and identify the source buyer in seconds.

  • Survives copy-paste, OCR and heavy paraphrasing
  • Anti-collusion: identifies groups of up to 3 buyers pooling their copies
  • A ~2,000-character excerpt is enough to trace
MEASURED, NOT PROMISED · 实测数据

Verified in production, not in a slide deck

100%
true-buyer identification across 15/15 regression tests
AUC 0.92–1.00
AI-paraphrase benchmark: 6 frontier models × 4 attack levels, 5,940 samples
90.7–100%
true-positive rate on AI-rewritten leaks at 1% false-positive (held-out)
AUC 0.94–0.97
full-length Chinese book paraphrase benchmark
606 / 606
lines traced in a print-scan-photo forensic examination of a full book
4 / 4
container attack classes caught (tamper / reorder / delete / truncate)
c = 3
colluders identified with zero false accusations
≥ 90%
codeword hit rate on verbatim excerpts of 8,000+ characters
HOW WE DIFFER · 与竞品的区别

Others watermark the file. We fingerprint the language.

Some competitors add a visible stamp showing the buyer's name; others (BooXtream, Custos) hide data inside the e-book file. Both approaches share one fatal weakness: the mark lives in the file, not in the text. Copy the words out — and the mark is gone. Our fingerprint is the wording itself: statistical patterns of synonym choice that are unique to each buyer and recoverable from any excerpt, from any device, in any format.

CapabilityInsightBridge Fingerprint™Visible watermark toolsFile-level watermark (BooXtream · Custos)Hard DRM (Locklizard · Adobe)
What is markedThe language itself (semantic layer)The page imageThe e-book fileThe e-book file
Survives copy-paste of text◐ (copy blocked, not traced)
Traces a leaked excerpt✓ — even ~2,000 characters✗ — needs the whole file
Survives print → scan / phone photo✓ — tested on real printouts
Survives AI paraphrasing (LLM rewrite)✓ — AUC 0.92–1.00 across 6 models, 5,940 samples
Anti-collusion (2–3 buyers pool copies)
Encrypted, tamper-evident delivery
Tamper / reorder / truncation detection✓ (4/4 attack classes)
Forensic evidence chain (court-ready)✓ — Merkle-sealed
Reader experienceAny device, any app, no plug-inVisible stamp on every pageAny deviceLocked to vendor reader app
Original manuscript retentionNone — destroyed after renderingVendor-held filesVendor-held filesVendor-held files
Price for an independent publisherFrom $169 per manuscript, one-timeCheap — but cosmetic≈ €0.60 / file + monitoring extra$500 / month, forever

Sources: vendor pricing pages & industry press, August 2026 (Locklizard, Vitrium, Digify, EditionGuard, BooXtream, Custos). Feature assessment based on public documentation.

PRICING · 定价

15–20% below comparable protection — and it never bills you twice

Hard-DRM suites charge $500/month, every month, and the protection ends when you stop paying. We charge once per manuscript: the fingerprint and the evidence chain last for the life of the edition.

Locklizard Safeguard$500 / monthHard DRM, visible dynamic watermark
Vitrium Securityfrom $500 / monthHard DRM
Digify$130 / monthDocument tracking, visible watermark
EditionGuard (Adobe DRM)$120 / month + $0.49–1.99 / downloadHard DRM / social DRM
BooXtream≈ €0.60 / fileFile-level invisible watermark only
InsightBridge Fingerprint™from $169 / manuscript — one-timeDual layer: encryption + semantic tracing
Up to 100k words

Short Work

$169$199
per manuscript
100k–300k words

Standard Monograph

$249$299
per manuscript
300k+ words

Major Work

$409$499
per manuscript
Per examination

Traceability Report

$39$49
per case
Buy in the Author Console →Launch pricing — 15–20% below list. Competitor prices from public vendor pages, August 2026.
PATENT PENDING · 专利申请中

A different science, not a different sticker

The core algorithms — dual-channel collusion-resistant fingerprint codes, order-bound container encryption, forensic-grade deterministic derivation, and distinguishing-bit consistency verification — are original work now in patent filing. This is why no other vendor offers excerpt-level semantic tracing: the math is different, not the marketing.

Protect your next manuscript

Register, pay, upload, and download your protected, traceable edition — self-service, in one sitting.

Open the Author Console