{"version":1,"type":"story","url":"https://digestai.news/story/openais-textgrain-watermark-fails-eu-tests-on-short-or-edited-text","json":"https://digestai.news/story/openais-textgrain-watermark-fails-eu-tests-on-short-or-edited-text.json","markdown":"https://digestai.news/story/openais-textgrain-watermark-fails-eu-tests-on-short-or-edited-text.md","slug":"openais-textgrain-watermark-fails-eu-tests-on-short-or-edited-text","headline":"OpenAI’s textGrain watermark fails EU tests on short or edited text","summary":"The EU’s AI Act requires AI-generated text to be watermarked for transparency, but OpenAI’s new **textGrain** system—designed to comply—struggles in real-world use. Tests show detection rates drop sharply for texts under 150 words or after editing, with synonym replacement cutting accuracy from 92% to 17%. OpenAI claims textGrain outperforms Google’s **SynthID Text** in its own tests, but those results lack independent verification. The EU’s rules demand watermarks be *effective* and *robust*, yet no clear benchmarks exist for what constitutes compliance. Critics like **Thomas Souverain** argue the system may not meet legal standards, risking fines of up to 7% of OpenAI’s revenue if found non-compliant. OpenAI has not responded to requests for comment, and its collaboration with researchers—including **John Thickstun** and the **Kempelen Institute**—remains in early stages. Souverain calls for open-sourcing textGrain to allow broader scrutiny, citing Meta’s **SynthID** as a model for transparency.","keyPoints":["OpenAI’s textGrain watermark detects AI-generated text at 95% accuracy for 400-token passages but drops to 80% for 200 tokens","Editing text reduces detection rates: replacing 10% of words cuts accuracy from 92% to 66%","EU AI Act lacks measurable thresholds for watermark effectiveness, leaving compliance uncertain"],"whyItMatters":"If textGrain fails EU scrutiny, OpenAI could face fines and reputational damage, while competitors gain an edge. For users, unreliable watermarks undermine trust in AI-generated content’s traceability, complicating compliance with emerging transparency laws.","category":{"slug":"policy","name":"Policy & Regulation","url":"https://digestai.news/category/policy"},"entities":{"companies":["OpenAI","Google DeepMind","Meta"],"models":["textGrain","SynthID Text"],"people":["Xiang Li","Thomas Souverain","John Thickstun","Dominik Macko"]},"firstPublishedAt":"2026-11-05T15:00:03Z","updatedAt":"2026-11-05T15:00:03Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"IEEE Spectrum AI","title":"AI Text Watermarking May Not Be Able to Meet the EU’s New Rules","url":"https://spectrum.ieee.org/ai-watermark-openai-textgrain-europe","publishedAt":"2026-11-05T15:00:03Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"OpenAI’s textGrain watermark fails EU tests on short or edited text\", 5 November 2026, https://digestai.news/story/openais-textgrain-watermark-fails-eu-tests-on-short-or-edited-text","publisher":"Digest AI","title":"OpenAI’s textGrain watermark fails EU tests on short or edited text","datePublished":"2026-11-05T15:00:03Z","url":"https://digestai.news/story/openais-textgrain-watermark-fails-eu-tests-on-short-or-edited-text"},"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"}