OpenAI to watermark ChatGPT text in the EU with textGrain
OpenAI will add an invisible watermark to ChatGPT and Codex text in the European Union over the coming weeks.
The technology is called textGrain. It works by subtly influencing the model’s word choices to create a hidden statistical signal that detection software can find.
The rollout responds to the EU AI Act, which requires generative AI providers to make text outputs identifiable in a machine-readable way.
OpenAI admits the system is not foolproof. Its own tests found detection dropped from 92 percent to 66 percent when 10 percent of words were replaced with synonyms.
Outside the EU, watermarking will not be the default. API customers worldwide can choose to turn it on for select models.
Anthropic already watermarks Claude’s text worldwide. OpenAI plans to open source the textGrain technology, but has not set a date.
Khabar 24h Editorial Desk — our explainers are prepared by the Khabar 24h editorial team using AI-assisted research tools, and every piece is reviewed by a human editor before publishing. We do not claim original reporting: our work is turning complex topics into simple, accurate summaries. Spotted an error? Write to contact@khabar24h.com — our corrections policy aims for same-day review.
More from this author →
Leave a Reply