Documentation & Warning

⚠️ Important Warning

This tool is for educational and research purposes only. By using this tool, you agree to use it ethically and legally. Do not use these techniques for malicious activities.

Usage Guidelines

  • Use for AI security research and testing
  • Test your own systems' robustness
  • Educational demonstrations only
  • Do not use for bypassing security measures
  • Responsible disclosure of vulnerabilities

Attack Categories

🔤 Homoglyph Attacks

Replace characters with visually similar Unicode glyphs from different scripts (Latin → Cyrillic, Greek) to bypass text filters and AI detection systems.

Example: paypal.com → рaypal.com (Cyrillic р)

👻 Invisible Character Attacks

Insert zero-width spaces, zero-width joiners, and bidirectional override characters that are invisible to humans but can break text processing pipelines.

Example: hello world → he​l​lo wo​rld

⌨️ Typo & Misspelling Attacks

Generate keyboard distance-based typos and phonetic misspellings that mimic real-world typing errors to evade keyword-based filters.

Example: the → teh, you → yuo

🔄 Semantic Attacks

Replace words with contextually similar synonyms to preserve meaning while evading keyword-based detection.

Example: good → excellent, important → crucial

⚡ Multi-Vector Combined Attacks

Chain multiple attack types sequentially for maximum evasion capabilities. Combines character, word, and sentence-level attacks.

Intensity Levels

Low:Minimal modifications, high readability (~20% replacement rate)
Medium:Balanced modifications (~50% replacement rate)
High:Aggressive modifications (~80% replacement rate)
Evasion:Maximum evasion (~95% replacement rate, multiple techniques)