06/10/2026
Share:
06/10/2026
Share:
BRUSSELS, Oct 6: OpenAI will begin adding invisible watermarks to eligible text generated by ChatGPT and Codex in the European Union over the coming weeks, as the company moves to comply with new transparency requirements under the EU AI Act.
OpenAI announced the move on October 5, saying its textGrain technology embeds an invisible statistical signal into the model’s word choices. A detector can then analyze a passage for the presence of the OpenAI watermark.
The EU’s AI Act requires providers of generative AI systems to make AI-generated content identifiable through machine-readable markings. The transparency provisions, including requirements for machine-readable marks on AI-generated or manipulated content, began applying on August 2, 2026.
Unlike a visible label, OpenAI’s watermark will not appear as a symbol or notice in the text. Instead, textGrain subtly changes the statistical pattern of the model’s word selection. OpenAI said it does not insert hidden characters, invisible spaces or special punctuation into the text.
However, OpenAI cautioned that the technology has significant limitations. Shorter passages are more difficult to identify, while editing, paraphrasing and translation can weaken the watermark. In its testing, replacing 10% of the words in a 400-token passage reduced detection from about 92% to 66%, while replacing 25% reduced it to 17%.
OpenAI also stressed that a positive detection does not establish that an entire passage was written by AI, nor does it indicate how much human editing or creativity was involved. It cannot identify the user, account or conversation associated with the text, and the absence of a watermark does not prove that a person wrote the material.
The company will initially restrict access to its text watermark detector to approved researchers and expert organizations. OpenAI said the limited rollout is intended to help evaluate the technology and improve its reliability before broader access is considered.
OpenAI said the watermarking will initially apply to eligible ChatGPT and Codex text generated in the EU across all plans. It will not become a global default at launch.
Separately, OpenAI said API customers worldwide can opt in to text watermarking for selected models, while the feature will remain off by default for API users. The company also plans to make textGrain open source so other developers and researchers can build on the technology.
The move expands OpenAI’s broader effort to establish provenance signals for AI-generated content. The company already uses Content Credentials and invisible watermarking for supported images and SynthID watermarking for supported audio.