THE HOOK
OpenAI is now quietly fingerprinting the words ChatGPT writes, as long as you're reading them in the EU.
THE GIST
OpenAI says it'll add an invisible watermark to ChatGPT and Codex text for EU users on all plans over the coming weeks, to meet the EU AI Act's transparency rules. It works by nudging word choices into a pattern a detector can spot. Developers anywhere can opt in via the API; it's off by default.
THE TAKE
This is compliance with a decent algorithm underneath. OpenAI's own tests show swapping one word in ten for a synonym drops detection from about 92% to 66%, so anyone with a thesaurus walks free. The real tell is the geography: if watermarking were good for users, it wouldn't need a border. Regulation flipped this switch.
OPINION
Anthropic went global with Claude watermarking two months ago and got a user backlash for its trouble, which explains why OpenAI is keeping this regional. The real story will be who ships the detectors, because a watermark nobody checks is just a very expensive comma.
- Story type: Important story
- Impact: Medium
- Confidence: High (official announcement)
- Audience: Consumers
- What changed: OpenAI is adding invisible text watermarks to ChatGPT and Codex output for EU users, with opt-in API support worldwide
- What to watch: Whether a public detector ships and whether OpenAI makes the watermark a global default
- Reading time: 1 minute
