The Education Insights
Advertisement
  • Home
  • Magazines
  • Awards
  • Education Insights
  • Press Release
  • About Us
  • Contact Us
No Result
View All Result
  • Home
  • Magazines
  • Awards
  • Education Insights
  • Press Release
  • About Us
  • Contact Us
No Result
View All Result
The Education Insights
No Result
View All Result
Soon AI generated Text to Have Watermarks

Soon AI generated Text to Have Watermarks

Soon, your AI-generated text content will have a watermark embedded in it. Monday, OpenAI announced in a blog post that the company will start adding an invisible watermark to text generated by ChatGPT and CodeX in the EU to comply with the new EU AI Act.

European Union’s AI Act, which took effect on 2nd August 2026, requires every AI company to mark AI-generated content in a way that other systems can identify.

The AI company mentioned that the watermark feature will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans, in the EU. However, developers using OpenAI’s API anywhere in the world can turn it on for select models starting Monday 5th October 2026. The feature is off by default. OpenAI also mentioned that it is not making text watermarking a global default at launch.

Now you might think that the watermark is an actual symbol, but it’s not; instead, it works by subtly shaping the model’s word choices, leaving a pattern that normal readers might not understand, but an AI detector can pick up. As it lives in the words themselves, it leaves a mark when the text is copied and pasted. OpenAI said the watermark doesn’t identify the user and has no meaningful change in its models’ performance with it switched on.

Now you might think, can these watermarks be removed by editing? OpenAI’s tests say yes. In one test, replacing 10% of words with synonyms dropped detection from about 92% to 66%. The company also explained the fact that short passages, math answers, and translated text are harder to detect.

The organization also cautioned that a missing watermark “does not prove human authorship.” The text could be too short or too heavily edited, or it could be generated from a different AI model.

OpenAI had already built this feature in 2024, but held off before releasing it due to concerns that users would switch over to rivals that don’t have any watermark. However, a few months earlier, Anthropic said it would watermark the text generated by Claude worldwide, which also opened the door for OpenAI to implement its watermark feature.

Previous Post

Global Education Impact Awards 2026: Celebrating the Visionaries Shaping the Future of Education

The Education Insights

A strong foundation is the true identity of a Education. When the roots are deep-valued and solid, the path becomes farther and faster. At Education Insights, we work with a vision to revolutionize the future with new possibilities and innovations.

Follow Us On

Recent News

8 Changes AI Has Brought to Outsourcing Businesses

8 Changes AI Has Brought to Outsourcing Businesses

How Innovation and Technology Make Life Easier

How Innovation and Technology Make Life Easier

AI

Why AI Could Be Your Best Career Bet During a Tech Downturn

Latest Magazines

Copyright © 2026 - The Education Insights. All rights reserved.

  • Contact Us
  • About Us
  • Advertise With Us
  • Privacy & Policy
No Result
View All Result
  • Home
  • Magazines
  • Awards
  • Education Insights
  • Press Release
  • About Us
  • Contact Us

© Copyright 2026, The Education Insights | All rights reserved.