OpenAI's new textGrain watermarking system, designed to comply with the EU AI Act's transparency requirements, faces challenges in reliably detecting AI-generated text. Tests indicate that the watermarks often fail to persist through text editing or in passages under 150 words. While the system aims for a 1% false-positive rate, its detection accuracy drops significantly with shorter texts and substantial edits, raising questions about its effectiveness in meeting the EU's legal benchmarks. AI
IMPACT OpenAI's text watermark may not meet EU AI Act standards, potentially hindering AI-generated text detection and compliance efforts.
RANK_REASON The article discusses a specific product (textGrain watermark) and its limitations in meeting regulatory requirements, rather than a new model release or fundamental research.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
