iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
OpenAI says it will begin adding an invisible statistical watermark to eligible ChatGPT and Codex text generated in the European Union over the coming weeks. The company is not making the feature a global default at launch: API customers worldwide can opt in for select models, with watermarking off by default. The mark can help identify some OpenAI-generated text, but it cannot establish who wrote or owns a passage, how much a person contributed, or whether the content is true.
What OpenAI announced—and where it applies
In an October 5, 2026 announcement, OpenAI said it would roll out text watermarking called textGrain to eligible ChatGPT and Codex text outputs in the EU. The rollout is expected to take place over the coming weeks; the announcement does not say that every eligible account already has it.
| Product channel | Availability announced | Default |
|---|---|---|
| ChatGPT and Codex | Eligible text outputs for users across all plans in the EU; rollout over the coming weeks after October 5, 2026 | Part of the announced EU rollout |
| OpenAI API | Customers globally can opt in for select models; the available model list may change | Off by default |
OpenAI says detector access will initially be limited to approved researchers and expert organizations, with applications reviewed case by case. Its Help Center explanation says the detector indicates whether it detects an OpenAI watermark, without revealing a user, prompts, or conversations. OpenAI’s public image and audio verification tools are separate; the announcement does not say that those tools check text.
Free tools Windows power users keep installed
One-click scans. No signup required.
How the text watermark works
textGrain is a statistical pattern in a model’s word or word-piece choices. OpenAI says the model subtly shifts those choices so a detector can look for the resulting pattern. It is not a hidden string or special symbol inserted into the output: copying and pasting text will not expose a watermark character, invisible space, or distinctive punctuation.
#1 Best Overall
Because the signal depends on patterns across the text, its usefulness varies with the amount and kind of material available. Short passages and writing with tightly constrained word choices give a detector less to work with.
How reliable is text watermark detection?
OpenAI’s 2026 evaluation reported the following results. These are company-reported measurements, not an independent real-world guarantee, and should not be generalized to every language, model, subject, length, or editing pattern.
Rank #2
| Test condition | OpenAI-reported result |
|---|---|
| Psychology passages of 200 tokens, at a target 1% false-positive rate | About 80% detection |
| Psychology passages of 400 tokens, at a target 1% false-positive rate | About 95% detection |
| Baseline test before synonym replacement | About 92% detection |
| 10% of words replaced with synonyms | Detection fell to 66% |
| 25% of words replaced with synonyms | Detection fell to 17% |
OpenAI also reported substantially lower detection for mathematics, where there is less flexibility in word choice. The company cautions that performance in ideal test conditions does not ensure reliable results in everyday use.
Short passages and code are special cases
OpenAI’s Help Center says the EU Code of Practice does not require watermarks in outputs under 200 tokens—about 150 English words—or in code snippets. That is distinct from whether a detector could identify a watermark: OpenAI says short text often lacks enough material for reliable detection, while code has fewer plausible next-word choices.
Rank #3
What a positive or negative result can tell you
A positive signal is limited evidence that an OpenAI system may have generated or processed some of the passage. OpenAI says, “A watermark does not measure human contribution.” The result does not quantify a person’s editing, judgment, or creativity, identify who used the system, or settle ownership, responsibility, or legality. It also says nothing about whether the passage is accurate or misleading.
A negative result is not proof of human authorship. OpenAI says short, edited, or translated text may evade detection, as may output from unsupported or older models or another provider. For the same reason, the watermark should not be treated as a plagiarism detector or definitive authorship test.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How the EU rules relate to OpenAI’s choice
OpenAI describes its rollout as a response to the EU AI Act and commitments under the EU Code of Practice. The legal framework is broader than textGrain: the European Commission says AI Act transparency obligations apply from August 2, 2026, including machine-readable marks that enable detection of AI-generated or manipulated content. It also describes duties to inform users when they interact directly with AI and disclosure obligations for deployers in specified cases.
The Commission’s June 10, 2026 Code of Practice announcement says the Code is voluntary and provides practical steps for meeting the Act’s obligations. It identifies clear labelling in key cases, including deepfakes and AI-generated or manipulated text published on matters of public interest. The Commission’s July 20, 2026 transparency-guidelines announcement explains the broader obligations and their scope.
Best Value
These rules do not mean the EU requires OpenAI to use textGrain specifically, nor that every AI-generated text must receive an identical watermark. OpenAI’s technical implementation and rollout choices should be distinguished from the law and the Commission’s voluntary practical guidance.
Why the detector is not public at launch
OpenAI says it is initially withholding public detector access because missed watermarks and false positives could cause harm. It has not announced broad public availability; access is restricted to approved researchers and expert organizations.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
