The Hindu reported that OpenAI will add an invisible watermark to eligible ChatGPT and Codex text in the EU over the coming weeks. OpenAI said it covers all EU plans and is not a global default at launch. Since October 5, API customers anywhere can opt in for select models. The company ties the step to Article 50(2) of the EU AI Act, which asks for machine-readable marking of AI-made text.
OpenAI said the mark is a pattern in word choices, found later with a secret key. In its tests, swapping 25% of words for synonyms cut detection to about 17%. For now, only approved researchers and expert organisations can use the detector. The Hindu does not say who counts as approved, how to apply, or what makes a user "eligible".
A mark that only the vendor and approved parties can read gives a buyer little to act on. A company cannot check it alone, edited text may lose it, and OpenAI says a missing mark proves nothing. At the 1% false-positive target, checking 1,000 unmarked documents would wrongly flag about 10. Buyers should ask who can run checks and what happens to staff-edited text.