OpenAI said Monday in a blog post that it will add an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act, according to TechCrunch and The Verge. The AI Act's transparency rules, which took effect August 2, require AI companies to mark AI-generated content in a way other systems can identify. OpenAI said the watermark will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans, but only in the EU. Developers using OpenAI's API anywhere in the world can opt in for select models starting now; it is off by default. The watermark is not a visible symbol but subtly shapes the model's word choices, leaving a pattern a detector can pick up, and it travels with copied-and-pasted text. OpenAI said it does not identify the user and saw no meaningful change in model performance. The company published a technical report for the method, called textGrain, co-written with researchers from the University of Pennsylvania and Yale. OpenAI's tests show editing can weaken detection: replacing 10% of words with synonyms dropped detection from about 92% to 66%. Detector access is initially limited to approved researchers and expert organizations. The move follows Anthropic's August announcement to watermark Claude text worldwide.
- Watermarking rolls out over the coming weeks to EU ChatGPT and Codex users on all plans; not a global default at launch
- API customers worldwide can opt in starting now; off by default
- Done to comply with EU AI Act transparency rules that took effect August 2
- In tests, replacing 10% of words with synonyms dropped detection from about 92% to 66%
- Detector access limited to approved researchers and expert organizations; OpenAI says a missing watermark does not prove human authorship
What it means for you
If you use ChatGPT or Codex inside the EU, text they produce will soon carry a hidden marker that detectors can read — invisible to you, but identifiable as AI-generated. For most people this changes nothing day to day; the text looks and reads the same. It matters mainly if you care whether AI-generated work can be flagged as such, or if you rely on it looking fully human-authored.
Try this
If you're in the EU and produce client-facing or published writing with ChatGPT, decide now whether you're comfortable with that output being detectably AI-generated — and note that OpenAI's own tests show editing roughly a tenth of the words sharply reduces detection.
Who should care
EU-based businesses and individuals using ChatGPT or Codex for text, plus developers deciding whether to opt in to watermarking via the API anywhere in the world.
Skip this if
You're outside the EU and don't use OpenAI's API, or you don't care whether your AI-generated text is detectable — the watermark is invisible and performance is unchanged.
Sources: TechCrunch AI, The Verge AI — read the original