Editorial illustration for OpenAI Adds Subtle Text Watermark to ChatGPT in European Union
OpenAI Adds Invisible Watermark to ChatGPT in EU
OpenAI said Monday it will start embedding an invisible watermark in text produced by ChatGPT and Codex for users in the European Union, a move tied directly to the EU AI Act. The law's transparency requirements kicked in on August 2, and they obligate AI companies to mark generated content so other systems can detect it. Rather than slapping a visible tag on outputs, OpenAI's approach works underneath the surface: the model nudges its own word choices in a pattern invisible to a human reader but readable by a detector built for the job. That pattern rides along with the text through copy and paste, which matters if the goal is catching AI writing after it's left the original chat window.
The rollout starts in the EU across all ChatGPT and Codex plans over the coming weeks. Developers using OpenAI's API elsewhere can switch the feature on for certain models starting today, though it's off by default and OpenAI isn't making it a global standard yet. The company says the watermark doesn't tie back to individual users and hasn't hurt model performance in testing. Alongside the announcement, OpenAI released a technical report on the method, co-authored with researchers from the University of Pennsylvania and Yale.
Why this matters
For developers building on OpenAI's API, this changes nothing yet, the watermark only applies inside ChatGPT and Codex for EU users. But it's a signal worth tracking. OpenAI picked the narrowest possible compliance path: a word-choice pattern invisible to readers, detectable only with the right tool, and limited to one jurisdiction.
That's a company testing the waters before deciding whether to commit globally. If you're building products that ingest or redistribute ChatGPT output, especially anything touching EU users, you should ask who holds the detector and under what terms, because a watermark nobody outside OpenAI can read is more of a compliance checkbox than a transparency measure. Researchers studying AI-generated text detection should also pay attention to how this interacts with existing watermarking schemes like those from Google's SynthID, since regulatory pressure is pushing multiple labs toward similar but incompatible solutions.
The real test comes when OpenAI decides whether to extend this to the API or keep it EU-only indefinitely, that choice will tell us more about their actual stance on transparency than the blog post does.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source
Related Topic OpenAI Developer: AI Agent "Swarms" Waste Tokens, Create "Coordination Tax
AI Tools & Apps
From Archives OpenAI adds optional AI watermark for API users, mandatory in EU
From Archives ChatGPT Ads test integrates product carousels into AI image loading screen
From Archives OpenAI consults mathematicians after AI missteps in math research
Research & Benchmarks
From Archives OpenAI Tests Visual Ads on Image Results, Labels Them Clearly
Business & Startups
Common Questions Answered
Why is OpenAI adding an invisible watermark to ChatGPT text in the European Union?
OpenAI is implementing the invisible watermark to comply with the EU AI Act's transparency requirements that took effect on August 2. The law obligates AI companies to mark generated content so that other systems can detect it, and OpenAI chose an invisible watermarking approach rather than visible tags on outputs.
How does OpenAI's invisible watermark work in ChatGPT and Codex?
The watermark works by having the model nudge its own word choices in a specific pattern that is invisible to human readers but detectable by systems with the right tools. This approach embeds the watermark underneath the surface of the generated text rather than adding visible markers.
Does the ChatGPT watermark apply to developers using OpenAI's API?
No, the invisible watermark currently only applies to text generated inside ChatGPT and Codex for users in the European Union. For developers building on OpenAI's API, this change does not affect their implementations at this time.
What does OpenAI's EU-only watermark strategy signal about future compliance?
OpenAI's decision to implement the narrowest possible compliance path—an invisible word-choice pattern limited to one jurisdiction—suggests the company is testing the waters before deciding whether to commit to global watermarking. This approach indicates OpenAI is carefully evaluating how to balance regulatory requirements with product design across different markets.
Further Reading
- OpenAI will start watermarking ChatGPT's text in the EU - TechCrunch
- OpenAI is adding text watermarking in ChatGPT and Codex - The Verge
- Provenance signals (Content Credentials, SynthID) in OpenAI-generated content - OpenAI Help Center
- OpenAI is adding invisible watermarks to ChatGPT and Codex for users in the EU - Digital Trends
- OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide - The Decoder
