Skip to main content
I
Uni
UNICODE
Tools/Watermark & Zero-Width Sanitizer

AI Invisible Watermark & Zero-Width Text Sanitizer

Scan, detect, and neutralize invisible zero-width tracking watermarks, AI fingerprints (ChatGPT/LLMs), Trojan Source bidi exploits, and hidden steganography.

Test Presets:
Input Document for Forensic Watermark Analysis
188 visible chars•204 UTF-8 bytes•8 invisible tokens

High Risk: Active Tracking Watermark / Steganography Detected

Contains zero-width binary patterns or high-risk bidi overrides commonly used by AI watermarking tools, corporate DLP leak detectors, or Trojan Source exploits.

Detected Invisible Codepoints Breakdown (3 Unique Types)

Zero-Width SpaceU+200B

Zero horizontal width; widely used for invisible watermarks & text fingerprinting.

×3
high risk
Zero-Width Non-JoinerU+200C

Prevents Arabic/Indic ligatures; commonly used in binary leak tracing.

×3
high risk
Zero-Width JoinerU+200D

Forces emoji combinations and ligatures; used as byte delimiter in steganography.

×2
high risk

Visual Anomaly Highlighter (Invisible Glyphs Rendered as Badges)

Red badges indicate characters invisible to human eyes
Artificial intelligence models[ZWSP] can embed[ZWNJ] invisible zero-width[ZWNJ] tracking tokens[ZWJ][ZWSP] into generated essays to trace content provenance[ZWNJ] and verify AI authorship without[ZWJ][ZWSP] visible alteration.

Watermark & Invisible Zero-Width Sanitizer

Detect, count, and strip invisible zero-width characters and AI watermarks using regex.

1# Python 3 - Detect & Strip Invisible Watermarks
2import re
3
4# Comprehensive Unicode pattern for zero-width & invisible format characters
5ZW_REGEX = re.compile(r'[​-‍⁠‎‏‪-‮­]')
6
7def sanitize_text(text: str) -> tuple[str, int]:
8 matches = ZW_REGEX.findall(text)
9 clean_text = ZW_REGEX.sub('', text)
10 return clean_text, len(matches)
11
12# Example: Suspicious text containing zero-width tracking watermarks
13raw_text = "ChatGPT\u200B generated\u200C text\u200D with hidden leak tracing."
14clean, count = sanitize_text(raw_text)
15
16print(f"Found {count} invisible characters!")
17print("Sanitized text:", clean)
Python 3Zero external runtime dependencies • Standard Library
UTF-8 & Unicode 16.0 Compatible

The Cryptographic Science of AI Watermarks & Zero-Width Steganography

Text Steganography is the practice of concealing secret information inside plain text without altering its visual layout or readability. In modern artificial intelligence systems (such as Large Language Model token generators) and corporate leak tracing tools, this is accomplished by translating secret binary sequences (0s and 1s) directly into invisible zero-width Unicode characters:

• Binary 0 → U+200B (Zero-Width Space)
• Binary 1 → U+200C (Zero-Width Non-Joiner)
• Byte Delimiter → U+200D (Zero-Width Joiner)
• Session Tracking Flag → U+FEFF (Zero-Width No-Break Space / BOM)

Because these characters occupy exactly 0 pixels on screen and paper, a document containing hundreds of tracking tokens looks 100% identical to the naked eye. However, when copied across email, Slack, Microsoft Teams, PDF, or text files, the invisible binary byte stream remains intact.

Cybersecurity, DLP & Whistleblower Forensic Applications

  • •Data Loss Prevention (DLP): Enterprises tag sensitive internal strategy memos with recipient employee IDs to pinpoint unauthorized press leaks.
  • •Whistleblower & Journalist Protection: Investigative reporters use our sanitization studio to strip hidden tracking markers from leaked government or corporate files before public disclosure.
  • •Trojan Source Defense: Detects and neutralizes malicious bidirectional overrides (U+202E, U+202D) that visually mask executable exploits in source code repositories.
  • •100% Client-Side Privacy: All watermark analysis, binary extraction, and purification occur locally in your browser memory. No text ever leaves your machine.

Frequently Asked Questions (FAQs)

How do AI models use zero-width characters to watermark generated text?+

Certain AI platforms and enterprise wrappers inject pseudo-random combinations of zero-width spaces (U+200B) and zero-width joiners (U+200D) between ordinary words. When an AI-generated essay or response is copy-pasted, this invisible signature acts as a cryptographic fingerprint proving AI origin.

Can social media platforms (Twitter/X, Facebook, LinkedIn) strip invisible watermarks?+

Some platforms sanitize non-essential invisible characters during post submission. However, pasting text into email, PDF, Slack, Microsoft Teams, Word, Google Docs, and code editors preserves zero-width payloads with 100% fidelity.

How does the "Purify & Strip" button sanitize my text?+

Our sanitizer uses an exhaustive Unicode Regular Expression scanning for all 17+ zero-width spaces, joiners, isolates, and Trojan Source bidirectional format overrides (U+200B through U+200D, U+FEFF, U+2060, U+202A through U+202E), cleanly dropping them while leaving your visible sentences completely untouched.

What are Trojan Source attacks and why are they dangerous?+

Trojan Source attacks exploit Unicode bidirectional override characters (such as U+202E Right-to-Left Override) in source code files. To a human code reviewer, the code appears innocent, but compilers and interpreters parse the characters in their true logical order, executing hidden malicious commands. Our scanner immediately flags these as Critical Risk.

Is any of my sensitive text sent to external servers or logged?+

No. iloveunicode operates 100% locally in your client web browser using the Web Standards TextEncoder and JavaScript string manipulation engines. Nothing is ever transmitted over the network.