OpenAI will add an invisible watermark to ChatGPT and Codex text in the EU. Its tests show editing can weaken it.
OpenAI will introduce an invisible watermark to qualifying ChatGPT and Codex texts within the European Union over the coming weeks. Their testing indicates that replacing certain words with synonyms can weaken the watermark’s effectiveness. In one test, substituting 25% of the words in 400-token English passages with synonyms lowered detection rates from approximately 92% to 17%.
From today, API users worldwide can opt to activate watermarking for select models, though it stays off by default. The text detection tool is not publicly accessible at launch; it is currently limited to approved researchers and expert organizations.
OpenAI states that the ChatGPT and Codex watermark will reach eligible users across all plans within the EU only. Initially, text watermarking won’t be a default feature globally.
The Oct. 5 update does not specify which models are part of the API opt-in or whether EU ChatGPT users can disable the watermark. The company says it is working with cloud partners to add watermarking to outputs from its models via their services in the coming weeks.
The transparency rules under Article 50 of the EU AI Act began applying on Aug. 2, 2026. The European Commission states that AI systems placed on the market before this date have until December 2 to meet the marking and detection obligation. The voluntary Code of Practice on Transparency of AI-generated Content provides organizations with an optional way to show their compliance.
Source link







