OpenAI will add invisible AI text watermarking to eligible ChatGPT and Codex output in the European Union over the coming weeks, making the company’s first regional default for identifying generated text. The announcement, made on 5 October 2026, links the rollout to the EU AI Act while acknowledging that text provenance technology is still developing and cannot reliably establish who wrote, edited or is responsible for a passage.
The move is part of OpenAI’s broader content-provenance strategy. In its announcement on EU text provenance, the company said watermarking will apply to eligible ChatGPT and Codex users across all plans in the EU. It will not be a global default at launch. For API customers worldwide, watermarking for selected models will be available as an opt-in feature and will remain off by default. OpenAI also said it is working with cloud partners to make watermarking available for model outputs accessed through their services.
OpenAI calls the technology textGrain. It adds an invisible statistical signal through the model’s word choices, which a specialised detector can assess. The company plans to update its technical report with further details and make the technology available as open source, but detector access will initially be limited to approved researchers and expert organisations.
OpenAI’s published evaluations show why that access is restricted. At a target false-positive rate of 1%, the detector identified watermarks in about 80% of 200-token psychology passages and about 95% of 400-token passages. Results were lower for mathematics, where there is less flexibility in word choice. Editing also weakened the signal: replacing 10% of words with synonyms reduced detection in one evaluation from about 92% to 66%, while replacing 25% reduced it to 17%.
The limitations matter for organizations that may want to use a detector as evidence in compliance, employment, education, content moderation or product-governance decisions. OpenAI states that a watermark does not measure human contribution, establish ownership or responsibility, identify a user, or verify accuracy. Conversely, the absence of a detected watermark does not prove that text was written by a person; content may be too short, translated, edited, generated by an unsupported model or produced by another provider.
The regional rollout follows the application of the EU AI Act’s Article 50 transparency obligations on 2 August 2026. The official Article 50 text requires providers of AI systems generating synthetic audio, image, video or text to ensure that outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, where technically feasible.
The European Commission says the requirement applies alongside separate obligations for deployers. Its Code of Practice on Transparency of AI-generated Content describes marking and detection measures for providers and disclosure requirements for certain deepfakes and AI-generated text published on matters of public interest. The code is voluntary, but the underlying Article 50 transparency obligations are legal requirements.
For organizations, the immediate issue is not simply whether a detector can classify a passage. Teams should distinguish provider-side marking from deployer-side disclosure, map where ChatGPT or Codex outputs enter products and business processes, and document how generated text is reviewed, edited and published. Nemko Digital’s overview of EU AI content-marking rules provides related regulatory context, while its EU AI Act resource outlines the wider risk-based framework.
The next developments to watch are OpenAI’s EU rollout, detector applications, the promised technical-report update and the planned open-source release. Organizations building a broader control environment can also place the change within established AI governance practices and ISO/IEC 42001 guidance, while treating watermark results as one provenance signal rather than conclusive proof of authorship or accuracy.