Transliteration Exploitation

Represents text from one script in another script's characters. The model understands transliterated text (Pinyin, Romaji, Cyrillic-to-Latin) but keyword classifiers looking for specific scripts miss it. Exploits the same multilingual gap as T2-AT-002 but at the character-representation level.

• Transliteration detection and normalization
• Multilingual keyword databases covering romanized forms

Chains with T2-AT-002 and T2-AT-004 for multi-vector evasion.