According to CoinMeta, OpenAI has announced that it will add invisible watermarks to ChatGPT in the European Union region to comply with EU AI legislation. These watermarks will be applied to chatbot texts and Codex encoded outputs. Tests by OpenAI have shown that if one-quarter of the words are replaced with synonyms, the detection rate drops to just 17%. The company acknowledges that the technology is not yet mature, so the detector is not currently available to the public. OpenAI also stated that the strength of the watermark diminishes with editing; for every 10 words replaced in a 300-word text, the detection rate drops from 92% to 66%. This system, named Textgrain, is designed to leave hidden statistical patterns through the model's word selection.