OpenAI said in an October 5 blog post that, to comply with the transparency rules of the EU AI Act that took effect on August 2, it will roll out an invisible text watermark to ChatGPT and Codex users on all plans in the EU over the coming weeks. Global API developers can enable the feature for select models starting immediately, with it off by default. The watermark shapes the model's word choices to create a detectable pattern without identifying users, and OpenAI also released a technical report, textGrain, co-written with researchers from the University of Pennsylvania and Yale.
EU government officials will gather in Brussels on Thursday to discuss how to craft global rules to address threats from new AI models, focusing on related work at the G7, G20 and the United Nations.