OpenAI will begin automatically watermarking ChatGPT outputs in the European Union starting in the coming weeks — the first time the company has enforced AI content labeling by default in any region. The move is a direct response to the EU AI Act, which took effect in August and mandates detectable AI disclosure for high-impact generative systems. Outside the EU, watermarking remains optional: users elsewhere won’t see it unless they manually enable it, and developers using the OpenAI API won’t get it turned on automatically at all. The watermark, called textGrain, is proprietary and works by subtly biasing word choices, patterns invisible to readers but detectable with a key and a specialized tool. OpenAI says it’s sharing that detector with a limited group of researchers and institutions, with others needing to apply for access. In testing, textGrain correctly identified watermarked text 92% of the time under ideal conditions. Long, untranslated prompts. But tampering breaks it fast: swapping just 10% of the output drops detection to ~63%. At 20%, it falls to ~17%. Shorter or translated text fares worse. That fragility mirrors earlier watermarking efforts from Google, Meta, and Anthropic, and points to why the EU AI Act’s requirement remains more symbolic than enforceable. Anthropic, for comparison, rolled out its own watermarking globally in August. Not just in the EU, and made it default across its Claude apps. OpenAI hasn’t said whether it plans broader rollout. For now, the policy draws a hard line: compliance where required, choice everywhere else. The rollout begins in the EU in the coming weeks.