ChatGPT will begin watermarking text generated in the European Union, joining Anthropic's Claude and Google's Gemini, which have implemented similar measures. This move comes in response to the EU AI Act's requirements for marking AI-generated content.
TL;DR
- ChatGPT introduces text watermarking for EU users, complying with the EU AI Act.
- Anthropic's Claude and Google's Gemini already employ text watermarks globally and in the app/API respectively.
- OpenAI, Anthropic, and Google use different methods and verification processes for their watermarks.
What happened
ChatGPT will introduce its textGrain watermark for text generated in the European Union, with no option to disable it. This follows the EU AI Act's requirement for providers of generative systems to mark synthetic content in a machine-readable format. The watermark will be available to all plans and API users can enable it worldwide from their organization or project settings.
Anthropic's Claude has been applying a worldwide text watermark since August 2026, using a version of Google's SynthID-Text. The watermark is applied to supported models and is verifiable by regulators, media, researchers, and educational bodies in a private preview.
Google's Gemini has been marking text generated in its app and on the web since May 2024, also using SynthID-Text. The watermark is applied to API and AI Studio text, but there is no public tool for text verification.
Why it matters
This development is significant for AI/ML developers and startup founders, as it highlights the increasing regulatory scrutiny on AI-generated content. The implementation of text watermarks can help ensure transparency and accountability in AI-generated text.
For tech investors, this move underscores the importance of compliance with regulatory requirements in the AI industry. Companies that fail to comply may face legal and reputational risks.
However, the effectiveness of text watermarks is still a matter of debate. OpenAI's textGrain has a detection rate of about 80% for 200-token passages and 95% for 400-token passages, with a 1% false positive rate. The watermark's effectiveness can vary depending on the language and the type of text.
Key facts
- ChatGPT's textGrain watermark will be introduced in the European Union, with no option to disable it.
- Anthropic's Claude applies a worldwide text watermark using a version of Google's SynthID-Text.
- Google's Gemini has been marking text generated in its app and on the web since May 2024, using SynthID-Text.
- OpenAI's textGrain has a detection rate of about 80% for 200-token passages and 95% for 400-token passages, with a 1% false positive rate.
- The EU AI Act requires providers of generative systems to mark synthetic content in a machine-readable format, with a deadline of August 2, 2026 for systems already on the market.
- About 190 organizations have signed the Code of Practice that accompanies the obligation, but each provider applies it in its own way and on its own schedule.
- Copilot, Mistral, and Meta AI have not documented any text watermarking as of October 2026.
Context
The EU AI Act is a comprehensive regulatory framework for AI systems, with the aim of ensuring that AI is developed and used in a safe, transparent, and accountable manner. The Act includes requirements for providers of generative systems to mark synthetic content in a machine-readable format, with the goal of ensuring that users can easily identify AI-generated content.
The implementation of text watermarks is a response to these regulatory requirements, as well as a growing concern about the potential misuse of AI-generated content. Text watermarks can help ensure that AI-generated content is transparent and accountable, and can be used to detect and prevent the spread of disinformation and other harmful content.
