OpenAI has begun rolling out textGrain — its invisible, machine-readable watermark for AI-generated text, to ChatGPT and Codex users in the European Union. The feature launches in the EU only, with no global rollout planned. It is not enabled by default for all users worldwide, and it does not apply to free-tier or non-EU accounts unless they opt in via API. Indian users of ChatGPT or Codex will not receive watermarked output unless they access the service through an EU-based account or use the API with the feature toggled on. And even then, only on select models. No Indian pricing, availability window, or local regulatory alignment is mentioned in the reporting. The system is being deployed specifically to comply with the EU AI Act’s requirement that generative AI outputs be identifiable in a machine-readable way. OpenAI says it will use the EU rollout as a learning phase before deciding whether, or how. To expand further. The watermark works by subtly adjusting word-selection patterns during generation. It survives copy-paste but degrades with editing: replacing 10% of words with synonyms cuts detection from ~92% to 66%. Replacing 25% drops it to 17%. Detection also falls sharply on short text (80% for 200-token psychology passages vs 95% for 400-token ones) and in domains like mathematics, where phrasing is more constrained. Translation further reduces reliability. The watermark carries no user ID, prompt data, or attribution, it signals only that the text was generated by a watermarked model. Its absence does not prove human authorship. OpenAI has made the detector available to researchers and expert organisations, and opened the watermark as an opt-in for API customers globally. But only on certain models, and only starting today. The company stresses that real-world detection remains probabilistic, not forensic. No timeline is given for India or other markets. The next step is feedback from EU users and partners, and a decision on whether textGrain becomes more than a compliance checkbox.