{
  "id": 2360030,
  "title": "The homoglyph trick against similarity checking is documented - as not working",
  "url": "https://urgent.news/2026/08/21/the-homoglyph-trick-against-similarity-checking-is-documented-as-not",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-08-21T10:39:14.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/yotta-fish/the-homoglyph-trick-against-similarity-checking-is-documented-as-not-working-2fdi"
  },
  "original_language": "en",
  "account": "A recent discovery reveals a flaw in similarity checking software that exploits the use of homoglyphs, or characters that appear similar to their Latin counterparts. By inserting a Cyrillic \"о\" into a word, a string matcher fails to recognize the alteration, yet visually, the document appears unchanged. This technique, documented by vendors but demonstrated to be ineffective, raises concerns about the reliability of such tools. The vendor's documentation assures that lookalike characters are automatically converted back before scanning, thus allowing the similarity check to proceed. However, the flags generated by the software do not signify wrongdoing but merely require human review. The flags cover a range of issues, including hidden text, invisible quotation marks, white text on white backgrounds, and replaced characters. The replacement technique aims to disrupt similarity matching by inserting characters from other alphabets that resemble Latin letters. However, Turnitin's algorithms detect and correct for these substitutions, raising flags but not hindering the matching process. The technique fails to achieve its intended effect, resulting in a documented instance of a bypass that does not work as intended. While most authors are unlikely to use this method, it highlights the importance of thoroughly examining submitted files for hidden characters or alterations before relying on similarity checking tools.",
  "summary": "Drop a Cyrillic о into the middle of a word and a string matcher stops seeing the word, while the page looks identical. It is the kind of thing you would write in ten lines and it is exactly the kind of thing a tool might do to your file without telling you. The vendor documentation says it swaps those characters back out before scanning, and raises a flag on the way past. I went looking because…",
  "key_points": [
    "Homoglyph trick exploits similarity checking software",
    "Cyrillic о inserted into word, fails recognition",
    "Turnitin algorithms detect and correct substitutions"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}