Editorial illustration for OpenAI Adds Invisible Watermark to ChatGPT and Codex for EU Users
OpenAI Adds Invisible Watermark to ChatGPT Output
OpenAI is baking an invisible marker called textGrain into ChatGPT and Codex output, starting with users in the European Union. The system tags generated text in a way that's machine-readable but invisible to anyone reading it on screen, meant to let tools later flag a passage as AI-written without changing how the text looks or reads. OpenAI says internal testing found textGrain "matched or exceeded" rival methods, including Google DeepMind's SynthID for text, the same technology Anthropic adopted in August.
The EU-only rollout isn't an accident. Both OpenAI and Anthropic are responding to requirements under the EU's AI Act, which pushes companies toward labeling synthetic content. OpenAI has been explicit that this won't become a worldwide default anytime soon, framing the regional launch as a way to gather real-world data before deciding whether to expand it.
Alongside the EU consumer rollout, OpenAI is opening up an opt-in version for API customers globally starting today, letting developers choose watermarked output for select models rather than have it forced on them. The company is also pairing the feature with benchmark scores comparing watermarked and unwatermarked text performance.
OpenAI also included scores from AI benchmarks showing similar performance from watermarked and unwatermarked text. But it notes that textGrain “does not guarantee reliable detection,” and says text watermarks don’t verify accuracy, determine who owns the text, measure how much a human contributed, or prove human authorship.
Why this matters
The EU-first rollout isn't really about watermarking technology, it's about regulation. The AI Act's transparency requirements are pushing OpenAI to build detection tools for a specific regulatory market before deciding whether to ship them everywhere else. For developers building on the API, that's worth watching closely: if textGrain stays EU-only, you could end up with geographically inconsistent outputs depending on where a user's account is registered, which complicates anything that touches content provenance or compliance tooling across borders.
OpenAI's claim that textGrain "matched or exceeded" approaches like Google DeepMind's SynthID is also the kind of comparison that deserves scrutiny rather than acceptance, since we haven't seen independent benchmarks yet. Founders building detection or authenticity-verification products should treat this as a signal that watermarking is becoming a compliance checkbox rather than a universal safety feature, at least for now. The real test is whether OpenAI expands this beyond the EU once the regulatory pressure eases, or whether it stays a regional patch.
That decision will tell us more about OpenAI's actual priorities than the announcement itself.
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 Alibaba's Qwen2.5 Debuts Over 100 Open-Source AI Models
From Archives OpenAI’s Sol Matches Astra Performance at a Fraction of the Cost
From Archives Wikimedia Links May Outage to 'Rogue' OpenAI Bots
Business & Startups
From Archives OpenAI Adds Subtle Text Watermark to ChatGPT in European Union
Common Questions Answered
What is textGrain and how does it work in ChatGPT and Codex?
textGrain is an invisible watermark that OpenAI has integrated into ChatGPT and Codex outputs, particularly for EU users. The system tags generated text in a machine-readable way that remains invisible to readers on screen, allowing tools to later identify content as AI-written without altering the text's appearance or readability.
How does OpenAI's textGrain compare to Google DeepMind's SynthID?
According to OpenAI's internal testing, textGrain matched or exceeded the performance of rival watermarking methods, including Google DeepMind's SynthID for text, which is the same technology that Anthropic has adopted. OpenAI also provided AI benchmark scores demonstrating similar performance between watermarked and unwatermarked text.
What are the limitations of textGrain watermarking according to OpenAI?
OpenAI explicitly states that textGrain does not guarantee reliable detection and notes that text watermarks cannot verify accuracy, determine text ownership, measure human contribution levels, or prove human authorship. These limitations highlight that watermarking is primarily a detection tool rather than a comprehensive verification system.
Why is OpenAI rolling out textGrain first to EU users?
The EU-first rollout is driven by the European Union's AI Act and its transparency requirements, which are pushing OpenAI to build detection tools for this specific regulatory market before deciding on broader deployment. This regulatory-driven approach means developers may face geographically inconsistent outputs depending on where a user's account is registered.
Further Reading
- OpenAI will start watermarking ChatGPT's text in the EU - TechCrunch
- OpenAI details new text watermarking system for ChatGPT, Codex, API - 9to5Mac
- OpenAI is adding invisible watermarks to ChatGPT and Codex for users in the EU - Digital Trends
- OpenAI starts watermarking ChatGPT and Codex text in the EU - The Next Web
- OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide - The Decoder
