🤖 OpenAI’s text watermark plan for the EU
On October 5, OpenAI announced plans to add invisible watermarks to eligible ChatGPT and Codex text in the EU over the coming weeks. The statistical signal is encoded in word choices, in response to EU AI Act requirements.
OpenAI reports no meaningful quality difference in its model benchmarks. Nevertheless, making generated text detectable introduces another consideration into how models choose their wording.
The signal can also weaken after editing. In a test of 400-token passages, replacing a quarter of the words with synonyms lowered detection from about 92% to 17%.
A detected watermark cannot quantify human input. Someone could write a passage themselves, have ChatGPT revise it, and receive text carrying a watermark. It does not establish how much of the final work was human-written.
Details: OpenAI
AI News · @aicontentum
October 9, 2026 246
