OpenAI is set to start embedding an invisible watermark in texts generated by its ChatGPT and Codex technologies in countries within the European Union (EU).The change aims to comply with transparency laws governing the use of artificial intelligence (AI) technology in EU countries, which came into effect on August 2, 2026.
The watermark will not be visible to the human eye as commonly understood. Instead, it involves a very small change in the way the technology selects words, which can only be detected by OpenAI’s AI-generated text detection program, called textGrain.
ALSO READ: https://agaciropress.com/imf-approves-35-7-million-for-rwanda-to-support-economic-growth/
The watermark has limitations
However, the watermark does not work with 100% effectiveness. OpenAI has shown that if a person changes only 10% of the words in a text, replacing them with words that have the same meaning, the detection rate can fall from 92% to 66%.
The watermark is also difficult to detect in short texts, texts translated into other languages, or mathematical answers. For this reason, only authorised experts will be given permission to use textGrain for now.
The development comes two months after rival company Anthropic announced plans to introduce a similar watermark in its texts globally, a move that did not please some users of the technology. OpenAI has had this method for some time but had refrained from using it because of concerns that doing so could lead to users abandoning its programs.
