{"version":1,"type":"story","url":"https://digestai.news/story/openai-announces-textgrain-watermark-for-eu-chatgpt-and-codex","json":"https://digestai.news/story/openai-announces-textgrain-watermark-for-eu-chatgpt-and-codex.json","markdown":"https://digestai.news/story/openai-announces-textgrain-watermark-for-eu-chatgpt-and-codex.md","slug":"openai-announces-textgrain-watermark-for-eu-chatgpt-and-codex","headline":"OpenAI announces textGrain watermark for EU ChatGPT and Codex","summary":"OpenAI introduced textGrain on October 5, 2026, an invisible watermark that embeds statistical patterns in the word choices of generated text. The feature can be turned on by API customers worldwide but is off by default, and OpenAI plans to apply it to ChatGPT and Codex outputs in the European Union in the coming weeks as a response to the EU AI Act.\n\nOpenAI’s internal evaluation shows the detector works best on longer English texts. For a 400‑token passage, replacing 10% of the words with synonyms lowered detection from roughly 92% to 66%, and a 25% replacement cut it to 17%. Under a 1% false‑positive target, detection was about 80% for 200‑token psychology texts and 95% for 400‑token texts, but rates fell for domains with limited word choice, such as mathematics. The company stresses that a detected watermark only indicates that its system was involved; it does not reveal the amount of human editing, ownership, or accuracy, nor does the absence of a watermark prove human authorship.\n\nOpenAI is limiting detector access to approved researchers and specialized organizations while layering text provenance with other signals like Content Credentials and SynthID for images and audio.","keyPoints":["textGrain adds invisible statistical patterns to word choice; optional in API, default off","Detection drops from ~92% to 66% with 10% synonym changes, to 17% with 25% changes in 400‑token text","Watermark presence only signals possible OpenAI involvement; not definitive proof of AI authorship"],"whyItMatters":"Provides a provenance clue for EU compliance and content verification, but its limits mean it cannot serve as a definitive AI detector.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["OpenAI"],"models":["ChatGPT","Codex"],"people":[]},"firstPublishedAt":"2026-10-17T17:22:00Z","updatedAt":"2026-10-17T17:22:00Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"note.com","title":"Invisible Watermarks in AI Text: What OpenAI's textGrain Can and Cannot Tell Us","url":"https://note.com/nagomi97/n/n058a301728c6?hl=en","publishedAt":"2026-10-17T17:22:00Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[],"thread":{"title":"AI Giants Adopt Invisible Text Watermarks","url":"https://digestai.news/thread/anthropic-s-invisible-text-watermarking-in-claude-ai","storyCount":7},"cite":{"text":"Digest AI, \"OpenAI announces textGrain watermark for EU ChatGPT and Codex\", 17 October 2026, https://digestai.news/story/openai-announces-textgrain-watermark-for-eu-chatgpt-and-codex","publisher":"Digest AI","title":"OpenAI announces textGrain watermark for EU ChatGPT and Codex","datePublished":"2026-10-17T17:22:00Z","url":"https://digestai.news/story/openai-announces-textgrain-watermark-for-eu-chatgpt-and-codex"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}