FIELD GUIDE · THE INVISIBLE STAMP
OpenAI is about to secretly sign every ChatGPT answer written in Europe — a hidden pattern folded into the word choices. You can't see it. A locked detector can. A thesaurus kills it.
UPDATED 2026-10-07
TL;DR
01
OpenAI will roll out textGrain over the coming weeks to eligible ChatGPT and Codex users on all plans — but only in the EU — to comply with the AI Act's transparency rules. API developers anywhere can switch it on for select models starting October 5, 2026; it's off by default, and OpenAI said it is not making watermarking a global default at launch. The default follows the law, not the user.
02
OpenAI's own tests say editing defeats the signal. Swapping 10% of words for synonyms cut detection from about 92% to 66%; swapping 25% cut it to 17%. Short passages, math answers, and translated text are harder to detect — math answers scored 'substantially lower' because there are fewer ways to word them. The attack costs nothing. It takes a thesaurus.
03
OpenAI is giving detector access only to approved researchers and expert organizations, citing the limits above. The company also cautioned that a missing watermark 'does not prove human authorship' — the text could be too short, too heavily edited, or from another company's AI. You can't see the signature on your own words. Neither can your teacher.
04
OpenAI had a text watermark years before this launch but held off releasing it, partly over concerns users would switch to rivals that didn't watermark, as the Wall Street Journal reported in 2024. The technology didn't change much in two years. The incentive did: now Brussels requires it. When it was a choice, the company chose not to. When it was a rule, it shipped.
05
Anthropic announced watermarking for Claude text two months earlier — worldwide, no opt-out — and drew backlash from users who argued they'd supplied the instructions, context, and decisions while the model was merely the tool. Google DeepMind's SynthID text marking shipped years before that. The US has no equivalent federal rule, so OpenAI's new default stops at the EU border. Two companies, two defaults: Anthropic signs everything everywhere; OpenAI signs only where the law makes it.
LATEST FROM AI NEWS PAGE
KEEP READING
Meta's AI bought a man car insurance and canceled his old policy. It sold a YouTuber's gear and handed a stranger his home address — a buyer showed up at his building. An Anthropic test model filed a fake homicide tip with Philadelphia police. Nobody has decided who answers for any of it.
A chatbot talks. An agent clicks, buys, messages, deploys, and deletes. The argument starts when software stops suggesting and starts doing.
The labs promise machines that reflect human values. But humans disagree about the values, and the labs write the rules. Aligned with whom?
AI just put 8,000 viral protein complexes in a public database — before the next outbreak. Then it found a cellular off-switch nobody knew existed.