OpenAI to add invisible watermarks to ChatGPT and Codex text in EU
By Automate Basics, written with AI from TechCrunch's original
OpenAI plans to add invisible watermarks to text from ChatGPT and Codex for eligible users in the EU. The marks may help identify AI-generated text, but editing can make them harder to detect.
What changed
OpenAI says text generated by ChatGPT and Codex for eligible users in the European Union will carry an invisible watermark. According to TechCrunch, the move responds to EU AI Act rules requiring AI-generated content to be marked so other systems can identify it.
The mark is a pattern in the model’s word choices, not a visible symbol. OpenAI says it remains in the text when someone copies and pastes it, and does not identify the user. Detecting the pattern requires more than simply reading the passage. OpenAI also says its testing found no meaningful change in model performance when watermarking was enabled.
Who can use ChatGPT and Codex watermarks?
OpenAI says the watermark will reach eligible ChatGPT and Codex users on all plans in the EU over the coming weeks. The announcement does not give an exact date for each user or specify what makes a user eligible. It also does not state a separate price for watermarking.
Developers using OpenAI’s API can choose to enable the watermark for select models from anywhere in the world, according to the company. That option starts off by default, and OpenAI is not making text watermarking a global default at launch. For a workplace outside the EU using ChatGPT or Codex directly, the announcement does not describe a comparable rollout.
How do I try it at work?
A longer ChatGPT draft for an internal memo is a sensible first task for an eligible EU user. Review the draft for accuracy, tone and missing context as usual. The watermark is invisible, so there is no mark to look for while editing or after pasting the text into another document.
Do not treat a detector result as a verdict on who wrote a passage. OpenAI says editing and translation can make its watermark harder to find, while short passages and math answers also pose problems. Detector access is initially limited to approved researchers and expert organizations, so most workers cannot check their own drafts with it. A missing mark does not establish that a person wrote the text, and a detected mark cannot show how much human judgment went into the final version.
Frequently asked questions
Will every ChatGPT user get text watermarking?
No. OpenAI says the ChatGPT and Codex rollout is for eligible users in the EU on all plans, over the coming weeks. It is not making watermarking a global default at launch. The announcement does not give an exact rollout date for each user.
Can editing remove a ChatGPT text watermark?
Editing can make the watermark harder to detect, according to OpenAI’s tests. Translation, short passages and math answers also create detection problems. A missing watermark therefore does not prove a person wrote the passage, and a detected watermark does not measure the person’s contribution.
Related reading
Anthropic offers eligible startups a free year of Claude TeamAnthropic has expanded Claude for Startups with a free year of Claude Team for eligible companies, plus API credits for building with Claude. Startups can apply to use the team plan and development benefits.
Mirror Particle builds AI model to predict consumer behaviorMirror Particle is building an AI model to predict how consumer behavior changes and explain the reasons behind its predictions. The company is targeting brands making market research, product and marketing decisions.
Google adds Trello card creation to Google Chat conversationsGoogle says Trello for Google Workspace now lets people turn Google Chat messages into Trello cards, alongside its existing email feature. The updated app is available to Google Workspace customers and people with personal Google accounts who also have Trello accounts.
Tools in this piece
Written with AI from TechCrunch's original and published after automatic checks: every figure here appears in the original, and no sentence is copied from it. The picture is AI-generated. The original is the authority.
Source: OpenAI will start watermarking ChatGPT’s text in the EU, TechCrunch, 5 Oct 2026.