{
  "id": 12682680,
  "title": "Invisible watermark: ChatGPT starts marking its text in the EU",
  "url": "https://urgent.news/2026/10/07/invisible-watermark-chatgpt-starts-marking-its-text-in-the-eu",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-07T18:20:49.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/getpack/invisible-watermark-chatgpt-starts-marking-its-text-in-the-eu-57ed"
  },
  "original_language": "en",
  "account": "OpenAI has launched an invisible watermark for text generated by ChatGPT and Codex in the European Union, as announced on October 5, 2026. The watermark, named textGrain, will appear over the coming weeks for eligible users on all plans, but it is off by default for API customers worldwide. OpenAI claims that textGrain matches or exceeds the performance of Google DeepMind's SynthID for text, and they plan to open-source the technology. The watermark is statistical, subtly shaping the model's word choices, and is undetectable on the page. OpenAI's own evaluation shows that the watermark detects about 80% of 200-token passages and 95% of 400-token passages, but it fails when the text is rewritten or translated. The watermark provides no evidence of AI-generated text for fraud, as Article 50 of the AI Act obliges providers to mark their outputs in a machine-readable format for EU users since August 2, 2026. Anthropic has been watermarking Claude's text worldwide since August 2026, and they have been compliant with the EU Code of Practice since July 2026. The watermark has structural weaknesses, being ineffective on factual answers, formulas, and code, and it is less reliable across different languages.",
  "summary": "First published in GetPack Magazine , sourced guides to use AI well as a student. You ask ChatGPT for a paragraph, you paste it into your document, and nothing changes on screen. Yet within a few weeks, that text will carry a mark in the European Union that you will never see. Here is how it works, what it proves, and the four things it does not. In short: OpenAI published a post titled « Our…",
  "key_points": [
    "OpenAI introduces invisible watermark \"textGrain\" for ChatGPT and Codex in EU",
    "textGrain matches or exceeds performance of Google DeepMind's SynthID",
    "Watermark is statistical, undetectable on page but detects 80-95% of passages"
  ],
  "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."
}