OpenAI has announced that it will start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the EU AI Act, OpenAI announced on Twitter. „In the EU, we'll start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the EU AI Act,“ OpenAI announced on Twitter.
Tagging will only be enabled for users located in the EU. In other countries, including the United Arab Emirates, Saudi Arabia, and the rest of the Middle East, OpenAI said the functionality will not be enabled automatically. Enterprise customers using the OpenAI API will be able to enable tagging independently, but companies around the world currently have no plans to implement regional rollout outside the EU.
How does an unreadable tag work?
OpenAI uses a system called „textGrain,“ which embeds an „invisible statistical signal“ into the model’s word choices. This signal doesn’t cause any visible changes to the text, but it allows specially designed tools to recognize that the content was generated by artificial intelligence. According to OpenAI, this method shouldn’t affect the quality of the text – „Across the benchmarks we use to assess Astra, our latest frontier model, we do not see meaningful performance differences with and without watermarking.“
However, some critics have raised concerns about the potential impact on text quality. They argue that inserting an invisible template could limit the model, especially when it comes to short answers where word count is limited. Additionally, editing can „weaken“ the tag signal, and AI detectors are still not perfect, which can lead to false positives.
Regional approach vs. global application
OpenAI’s choice to implement the tag only in the EU differs from other companies, such as Anthropic, which also introduced the tag in 2024 but admitted that it currently has no way to limit its regions. OpenAI says the regional approach allows it to „learn from real-world usage and feedback,“ allowing it to assess how the tag works in practice before making broader decisions.
Cooperation with researchers and regulators
OpenAI is inviting selected researchers and expert organizations to test and evaluate the tag detector. The company notes that „we will adapt our approach as evidence, standards, and regulatory requirements evolve,“ and plans to expand the detector’s availability once it is confident that the results can be interpreted responsibly.
What does this mean for users?
ChatGPT users in the European Union will soon notice that their query responses will be marked with an invisible mark. This will not affect the content or style of the text, but will provide a way to distinguish AI-generated content from human-written content. In other countries, this process will not be automatically enabled, but companies using the OpenAI API will be able to independently enable this feature if their business needs require it.
Future prospects
OpenAI plans to monitor how the tagging works in real-world use and make decisions about potential rollout outside the EU based on the data. There is currently no information on wider regional or global implementation, but the company’s openness to collaborating with researchers suggests that it is committed to creating a robust and regulatory-compliant solution.
However, it is important to note that Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in 2025 for alleged copyright infringement in the process of collecting training data. This fact shows that the development of AI technologies faces not only technical but also legal challenges.
OpenAI's move to introduce an unreadable tag in the EU is an important step towards transparency and regulatory compliance, but its practical impact will still remain to be tested and evaluated by both consumers and regulators.
Sources
- PCMag.com - OpenAI to Add Watermarks on ChatGPT's Text Outputs, But Only in the EU
- Online-tech-tips.com - ChatGPT Ads New Format: OpenAI Tests Visual Ads Inside Image Generation
