OpenAI will start watermarking ChatGPT’s text in the EU
OpenAI announced it will embed an invisible watermark in text generated by ChatGPT and Codex for users in the European Union, a step taken to meet the EU AI Act’s transparency requirements that came into force on August 2. The watermark, termed “textGrain,” subtly influences the model’s word choices to create a detectable pattern that persists when the text is copied or pasted, though it does not reveal the individual user’s identity. rollout will begin in the coming weeks for all EU‑based ChatGPT and Codex accounts across all subscription tiers, while developers worldwide can enable the feature via the API on a per‑model basis, though it remains off by default outside the EU.
The watermark’s design relies on a secret key that nudges thousands of next‑word predictions, allowing detectors to identify AI‑generated content without altering the visible text. OpenAI’s internal tests show the method does not degrade model performance, but it can be partially obscured: substituting roughly 10 % of words with synonyms reduced detection rates from about 92 % to 66 %. Short excerpts, mathematical answers, and translated passages also pose detection challenges, prompting OpenAI to limit initial detector access to vetted researchers and expert organizations. The company cautioned that the absence of a watermark does not guarantee human authorship, as heavily edited AI‑generated text or output from other AI systems may lack the mark.
OpenAI’s move follows Anthropic’s recent decision to watermark its Claude model worldwide, a policy that sparked user backlash over concerns of diminished attribution for human input. While OpenAI had previously developed a watermark, it delayed deployment over fears of user migration to non‑watermarked rivals. The new EU‑focused rollout aligns OpenAI with other major AI firms—Google, Meta, Microsoft, and Anthropic—that have pledged to adhere to the EU’s code of practice on AI‑generated content, signaling broader industry compliance with emerging regulatory standards.
⚡ Effects Interpreter
🌍World Economy
- ▶World markets have a habit of reading between the lines of stories like this.
- ▶Global commerce typically shrugs off modest shocks, but keeps one eye open.
🏙️Local Economy
- ▶The cost of a weekly food shop can creep up in the background.
- ▶Local wages and hours worked might bend slightly with the wider trend.
🏦Rates & Banks
- ▶Banks generally prefer a wait-and-see approach before touching their rates.
- ▶Rate-setters usually take their time digesting this kind of news.
❤️Health
- ▶Unsettling news can weigh on sleep and mood, so peace of mind matters.
- ▶Health anxieties triggered by this kind of news usually ease once facts replace speculation.
💷Wealth
- ▶Short-term wobbles like this tend to even out given enough time.
- ▶Savings and portfolios can see short-lived ups and downs after a story like this.
🏠Housing
- ▶Asking prices in the area might firm up or soften only gradually.
- ▶Property values usually need more than one headline to really move.