OpenAI enables EU-only watermark for ChatGPT outputs by default
OpenAI will automatically watermark text generated by ChatGPT in the European Union starting in the coming weeks. The move complies with the EU AI Act, which requires detectable AI-generated content. Outside the EU, the feature will remain optional and off by default. OpenAI’s proprietary method, called textGrain, embeds detectable patterns in word choices without altering readability. The…
Key points
- OpenAI will watermark ChatGPT outputs by default only in the EU to comply with the AI Act
- textGrain detection works 92% of the time but drops sharply with text edits or translations
- API users can still opt out, and detectors will be restricted to approved researchers
Anthropic introduced similar watermarking globally in August, but OpenAI’s approach is region-specific. The EU’s stricter stance contrasts with other regions, where watermarking remains voluntary. Critics note existing standards like SynthID and C2PA are also evadable, suggesting textGrain’s reliability may be limited.
The story so far
6 episodes →- OpenAI enables EU-only watermark for ChatGPT outputs by defaultthis story
OpenAI will watermark ChatGPT outputs by default—but only in the EU
Ars Technica AI · 6 October 2026
Loading the full article…
This text was published by Ars Technica AI and written by Samuel Axon. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- OpenAI and Anthropic urge UK to adopt binding AI rules · 2 src
- US regulators question banks on agentic AI, Chinese models and safeguards · 1 src
- Leading the Future PAC may use shell companies to hide AI campaign spending · 1 src
- AI’s refusal often fails: perils of AI safety · 1 src
- SemiAnalysis reports Chinese AI labs rarely publish safety results · 1 src
Comments
via GitHub Discussions