OpenAI to Watermark ChatGPT Text in European Union Under New AI Rules

OpenAI will begin adding an invisible watermark to text produced by ChatGPT and Codex in the European Union, TechCrunch and The Verge reported on Monday. The decision is intended to ensure compliance with mandatory transparency rules under the European Union AI Act, which took effect on August 2. [1][4]

According to TechCrunch, the watermarking system will roll out over the coming weeks to eligible ChatGPT and Codex users on all service plans within the European Union. Global developers using the company's application programming interface can enable the feature for select models starting today, although it remains turned off by default. TechCrunch added that OpenAI warned that editing text can make these hidden marks harder to detect. [1]

The Verge reported that OpenAI is calling the technology textGrain and claims it matches or exceeds existing watermarking approaches such as Google DeepMind's SynthID. While the watermarking feature is launching in the European Union to satisfy regional regulations, OpenAI stated that it will not immediately serve as its global default standard. [4]

In a separate announcement, The Verge and PYMNTS reported that OpenAI will start testing new visual advertisements inside ChatGPT in the United States later this month. The new ad format will display images of sponsored products and services to users when they generate images using the artificial intelligence tool. [3][6]