\n<\/p>\n
OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act, the company said Monday in a\u00a0blog post<\/a>.<\/p>\n The EU AI Act\u2019s transparency rules<\/a>, which took effect on August 2, require AI companies to mark AI-generated content in a way other systems can identify.<\/p>\n OpenAI said the watermark will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans, but only in the EU. Developers using OpenAI\u2019s API anywhere in the world can turn it on for select models starting today; it\u2019s off by default. OpenAI said it is not making text watermarking a global default at launch.<\/p>\n The watermark is not an actual symbol, but works by subtly shaping the model\u2019s word choices, leaving a pattern readers can\u2019t see, but a detector can pick up. Because it lives in the words themselves, it travels with the text when it\u2019s copied and pasted. OpenAI said the watermark doesn\u2019t identify the user, and that it saw no meaningful change in its models\u2019 performance with it switched on.<\/p>\n OpenAI also published a technical report<\/a> for its method, called textGrain, alongside the announcement. Co-written with researchers from the University of Pennsylvania and Yale, it walks through an example of using a secret key to sort next-word predictions to finish the sentence. Add hundreds of these nudges together, and the detector can spot AI-generated content using only the text and the key.<\/p>\n Can the watermark be removed by editing? OpenAI\u2019s tests suggest yes. In one test, replacing 10% of words with synonyms dropped detection from about 92% to 66%. The company also said short passages, math answers, and translated text are harder to detect.<\/p>\n \u201cThese limitations contribute to our decision to provide initial detector access only to approved researchers and expert organizations, who can help us evaluate reliability and responsible uses,\u201d said the company.<\/p>\n OpenAI also cautioned that a missing watermark \u201cdoes not prove human authorship.\u201d The text could be too short or too heavily edited, or it could come from another company\u2019s AI. <\/p>\n \u201c[Watermarks] can indicate that an OpenAI system generated or processed part of a passage, but not how much human judgment, editing, or creativity went into it,\u201d the company said.<\/p>\n The announcement comes two months after\u00a0Anthropic said it would watermark<\/a>\u00a0text generated by Claude, a move it\u2019s applying worldwide. That decision\u00a0drew backlash from some Claude users<\/a>, who argued they had supplied \u201cthe instructions, context, decisions\u201d while Claude was just \u201cthe tool.\u201d<\/p>\n OpenAI had built a text watermark before but held off on releasing it, partly over concerns that users would switch to rivals that didn\u2019t watermark,\u00a0The Wall Street Journal reported<\/a>\u00a0in 2024.<\/p>\n Anthropic, Google, Meta, Microsoft, and OpenAI are among the companies that have committed to following the\u00a0EU\u2019s code of practice<\/a>\u00a0on AI-generated content.<\/p>\n<\/div>\n When you purchase through links in our articles, we may earn a small commission<\/a>. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n