OpenAI to Watermark ChatGPT Text in EU for Compliance

OpenAI's Watermarking Initiative for ChatGPT in the EU

OpenAI has announced that it will begin watermarking text generated by ChatGPT and Codex in the European Union. This move is in response to the EU AI Act's transparency requirements, which took effect on August 2, 2026. The watermarking feature will be rolled out in the coming weeks for eligible users in the EU, while developers using OpenAI's API can activate it for specific models starting immediately.

Mechanism of Watermarking

The watermarking process does not involve visible symbols; instead, it subtly alters the model's word choices to create a pattern that is undetectable to readers but can be identified by specialized detection systems. This method, referred to as textGrain, ensures that the watermark is embedded within the text itself, allowing it to persist even when the text is copied and pasted. OpenAI has stated that the watermark does not identify individual users and that its implementation has not significantly affected the performance of its models.

Technical Details and Research Collaboration

OpenAI published a technical report detailing the watermarking method, co-authored with researchers from the University of Pennsylvania and Yale. The report explains how a secret key can be utilized to influence next-word predictions, which, when aggregated, enables detection of AI-generated content. This approach highlights the technical sophistication behind the watermarking process and its reliance on advanced AI techniques.

Limitations and Detection Challenges

While the watermarking system is designed to be robust, OpenAI acknowledges certain limitations. Tests indicate that editing the text can reduce detection accuracy; for example, replacing 10% of the words with synonyms lowered detection rates from approximately 92% to 66%. Additionally, shorter texts, mathematical answers, and translations present further challenges for detection. OpenAI has decided to provide initial access to detection tools only to approved researchers and expert organizations to evaluate their reliability and responsible use.

Context within the AI Industry

This announcement follows a similar commitment from Anthropic, which stated it would watermark text generated by its AI model, Claude, on a global scale. The decision by Anthropic faced criticism from users who felt that their contributions were being undervalued. OpenAI's cautious approach to watermarking stems from previous concerns about potential user migration to competitors that do not implement such measures.

Implications for Content Creators

For creators and studios, the introduction of watermarking has significant implications. It enhances the traceability of AI-generated content, which may affect how such content is used in various media. As compliance with regulations becomes increasingly important, understanding the mechanisms of watermarking will be crucial for creators who rely on AI tools for content generation.

Future Considerations

OpenAI's watermarking initiative is part of a broader trend among AI companies, including Google, Microsoft, and Meta, to adhere to the EU's code of practice on AI-generated content. As regulations evolve, content creators must stay informed about these developments and consider how compliance measures may influence their workflows and the perception of AI-generated content in the marketplace.

Conclusion

The implementation of watermarking by OpenAI represents a significant step towards transparency in AI-generated content. As the technology matures and regulations tighten, creators should prepare to adapt to these changes, ensuring that they understand both the capabilities and limitations of watermarking in their use of AI tools.