OpenAI adds watermarks to ChatGPT text in the EU to comply with AI Act
The move affects API customers globally, with opt-in available for select models. The EU AI Act mandates AI-generated text to be identifiable, prompting changes in how AI outputs are marked.

OpenAI has introduced a new text watermarking system to comply with the EU AI Act, which mandates that providers of generative AI systems make AI-generated text identifiable. This system allows API customers globally to opt in to watermarking for select models, starting today. The initiative is part of a broader effort to ensure transparency and accountability in AI-generated content, aligning with regulatory requirements across the European Union.
A few months ago, Anthropic faced controversy when it announced that Claude would begin watermarking text to comply with the EU AI Act. This sparked debate about the implications of such measures, particularly concerning the integrity of AI-generated content. OpenAI’s approach now mirrors this regulatory push, aiming to balance compliance with user expectations for clear and traceable AI outputs.
According to OpenAI, the watermarking system will be available to API customers globally, with opt-in options for select models. In evaluations, replacing 10% of words with synonyms in 400-token passages reduced detection rates from about 92% to 66%. This highlights the potential challenges in ensuring watermarking remains effective while maintaining the quality and usability of AI-generated text.
The introduction of watermarks may increase operational costs for AI providers, as additional layers of compliance and verification are required. It could also lead to vendor lock-in, as companies may become dependent on specific watermarking technologies. Governance frameworks will need to evolve to address these changes, ensuring that watermarking does not hinder innovation or create unnecessary barriers for developers and users.
As the EU AI Act continues to shape the regulatory landscape, the impact of watermarking on AI systems remains a developing issue. OpenAI’s move signals a broader industry trend toward compliance and transparency, but the long-term effects on AI deployment, user trust, and market dynamics are still unfolding.
Sources
- https://9to5mac.com/2026/10/05/openai-details-new-text-watermarking-system-for-chatgpt-codex-and-the-api/
- https://indianexpress.com/article/technology/openai-is-adding-watermarks-to-chatgpt-text-in-the-eu-10908898/
- https://www.gadgets360.com/ai/news/openai-invisible-watermarks-chatgpt-text-ad-format-testing-timeline-regions-12144879
- https://www.theverge.com/ai-artificial-intelligence/1004880/openai-chatgpt-text-watermarks-eu-ai-act