OpenAI says it is adding invisible text watermarks to eligible ChatGPT and Codex output in the EU in response to the EU AI Act. That is evidence that regulation prompted a concrete company response—not proof that the watermark reliably identifies AI writing or that the law has achieved its broader goals. The rollout was described as phased, and OpenAI’s Help Center now says EU ChatGPT-generated text carries textGrain; neither source establishes the status of every account as of October 7, 2026.
What OpenAI announced
On October 5, 2026, OpenAI announced an EU-focused plan to add invisible watermarks to eligible text generated by ChatGPT and Codex. The company said the consumer-product rollout would happen over the coming weeks and would not make text watermarking a global default at launch. Its Help Center’s present-tense description of EU ChatGPT text as watermarked may reflect rollout progress, but gives no account-by-account availability detail.
As an Amazon Associate I earn from qualifying purchases.
| Product or route | What OpenAI says | Availability qualification |
|---|---|---|
| ChatGPT and Codex in the EU | Invisible watermarks for eligible text output | OpenAI described a phased rollout over the coming weeks; exact availability for individual accounts is not stated. |
| OpenAI API | Text watermarking for select models | Available as a global opt-in for API customers; off by default. |
| Watermark detector | Detector access for approved researchers and expert organizations | Applications are reviewed case by case. OpenAI did not make the detector public at launch, citing the risk that false positives and missed marks could mislead users. |
These details come from OpenAI’s October 5 announcement and its Help Center entry, “Provenance signals in OpenAI-generated content.” The announcement describes the rollout plan; the Help Center’s wording does not establish that every eligible user had received the feature by October 7.
Free tools Windows power users keep installed
One-click scans. No signup required.
What the EU AI Act requires—and what it leaves open
Article 50(2) of the EU AI Act requires providers of systems that generate synthetic audio, images, video, or text to ensure that outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The law calls for technical solutions to be effective, interoperable, robust, and reliable as far as technically feasible, taking account of content-specific limits, implementation costs, and the state of the art. It does not name textGrain or prescribe one exclusive watermarking method. Article 50 also includes exceptions, including for systems whose function is assistive standard editing or that do not substantially alter input data or its semantics.
#1 Best Overall
| Duty | Who it applies to | What it concerns |
|---|---|---|
| Machine-readable marking and detectability | Providers of systems generating synthetic content | Marking outputs, subject to technical feasibility and the statute’s exceptions. |
| Disclosure of AI-generated or manipulated text | Deployers publishing such text to inform the public about matters of public interest | Disclosure is not required where the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for publication. |
The provider-side marking requirement and a publisher’s disclosure obligation are related but distinct. A watermark is not, by itself, a publisher’s disclosure, and a publisher’s disclosure does not establish that provider-side marking occurred.
The European Commission says these transparency obligations applied from August 2, 2026. Its Code of Practice on Transparency of AI-generated Content is voluntary; Article 50’s legal obligations are not. The Commission presents the code as a practical compliance route, with separate provisions for provider marking and detection and deployer disclosure.
How textGrain puts a signal into generated text
OpenAI describes textGrain as subtly changing the model’s randomized choices among possible words or word pieces. The pattern across a passage creates a statistical signal that a detector can test. It is part of the wording itself—not a visible label, a string of hidden characters, invisible spaces, or unusual punctuation.
In a technical report dated October 5, 2026, OpenAI and academic coauthors describe coupling token generation to keyed randomness through an optimal-transport problem, with costs based on Gumbel random variables and KL regularization. The method groups vocabulary tokens into blocks and uses an entropy budget to control the average amount of sampling randomness removed. A detector uses the text and a secret key to test for a statistical dependency. These are descriptions of the authors’ method and assumptions, not independent evidence of how well it performs in routine use.
Rank #3
What OpenAI’s published performance figures show
The following figures are from OpenAI’s October 5, 2026 announcement. They are company-reported evaluations, not independent replication or a general accuracy guarantee.
- At a target false-positive rate of 1%, OpenAI said its detector identified watermarks in about 80% of 200-token passages and about 95% of 400-token passages in content such as psychology.
- For mathematics, where there is less flexibility in word choice, OpenAI said detection rates were substantially lower but did not give a single percentage in the announcement.
- In 400-token passages, OpenAI reported that replacing 10% of words with synonyms reduced detection from about 92% to 66%; replacing 25% reduced it to 17%.
OpenAI also reported results for its Astra model on seven named benchmarks, saying it found no meaningful performance differences with and without watermarking. Examples from its announcement include Artificial Analysis Intelligence Index scores of 49.57 without watermarking and 49.76 with it, AutomationBench scores of 34.09% and 34.86%, and DeepSWE v1.1 scores of 72.80% and 71.68%, respectively. These are OpenAI-selected benchmark results, not independent validation of either model quality or watermark performance across other models and tasks.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What a detector result can—and cannot—establish
OpenAI says a positive result can indicate that an OpenAI system generated or processed some of a passage. It does not establish who used the system, how much a person contributed, who owns the text, who is responsible for publishing it, whether the text is accurate, or whether its use was lawful or required disclosure.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →A negative result is not proof of human authorship. Text may be too short, edited, translated, generated by an unsupported model, or produced before watermarking was applied. OpenAI describes text watermarking and detection as early technologies with significant limitations.
Best Value
- Length and genre matter: OpenAI’s reported figures differ between 200- and 400-token passages, and it reports lower detection for constrained writing such as mathematics.
- Editing can weaken the signal: OpenAI’s synonym-replacement results show reduced detection in its 400-token tests.
- Coverage matters: A detector result can speak only to signals and systems it supports; the sources do not establish universal coverage of AI-generated text.
- A signal is not a verdict: Watermark detection is not a truth test, a complete authorship history, or a measure of human contribution.
In what sense does regulation work here?
The narrow causal claim is supported: OpenAI explicitly framed its approach as a response to the EU AI Act, and the Commission says the transparency obligations had applied since August 2, 2026. Regulation therefore prompted or accelerated a specific company response: an EU-focused text-provenance rollout, alongside a separate opt-in path for API customers worldwide.
That does not show that the law requires textGrain specifically, that the watermark will reliably identify AI-generated text in ordinary use, or that watermarking alone reduces deception. The sources available for this article do not include an independent regulator or peer-reviewed evaluation of deployed textGrain performance. OpenAI says it may revisit its approach as technology, evidence, standards, and requirements evolve.
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.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
