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 aÃÂ blog post.
The EU AI ActâÂÂs transparency rules, which took effect on August 2, require AI companies to mark AI-generated content in a way other systems can identify.
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 OpenAIâÂÂs API anywhere in the world can turn it on for select models starting today; itâÂÂs off by default. OpenAI said it is not making text watermarking a global default at launch.
The watermark is not an actual symbol, but works by subtly shaping the modelâÂÂs word choices, leaving a pattern readers canâÂÂt see, but a detector can pick up. Because it lives in the words themselves, it travels with the text when itâÂÂs copied and pasted. OpenAI said the watermark doesnâÂÂt identify the user, and that it saw no meaningful change in its modelsâ performance with it switched on.
OpenAI also published a technical report 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.
Can the watermark be removed by editing? OpenAIâÂÂs 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.
âÂÂThese 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,â said the company.
OpenAI also cautioned that a missing watermark âÂÂdoes not prove human authorship.â The text could be too short or too heavily edited, or it could come from another companyâÂÂs AI.
âÂÂ[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,â the company said.
The announcement comes two months afteràAnthropic said it would watermarkàtext generated by Claude, a move itâÂÂs applying worldwide. That decisionàdrew backlash from some Claude users, who argued they had supplied âÂÂthe instructions, context, decisionsâ while Claude was just âÂÂthe tool.âÂÂ
OpenAI had built a text watermark before but held off on releasing it, partly over concerns that users would switch to rivals that didnâÂÂt watermark,àThe Wall Street Journal reportedàin 2024.
Anthropic, Google, Meta, Microsoft and OpenAI are among the companies that have committed to following theàEUâÂÂs code of practiceàon AI-generated content.
Topics
When you purchase through links in our articles, we may earn a small commission. This doesnâÂÂt affect our editorial independence.
