OpenAI launches invisible text marks in ChatGPT and Codex in the EU through «textGrain» technology

OpenAI has begun rolling out a new system in the European Union that adds invisible marks to text generated by ChatGPT and Codex, in an effort to increase transparency and help identify content created by artificial intelligence.

The new technology, called «textGrain», will be available to users in the EU from next month, the company announced on its blog on Thursday, saying the marks are invisible to the human eye, but can be detected by special software.

“The aim is to help people understand when they are reading content created by AI, and to prevent the misuse of our models for disinformation,” the company said.

OpenAI said the system works by making small, imperceptible changes to the text, such as changing the spacing between words or the order of punctuation, which can be detected by an algorithm.

The company added that the marks will not affect the quality of the text, and that they will be applied to all text generated by ChatGPT and Codex in the EU, including in the free and paid versions.

However, OpenAI warned that the system is not foolproof, and that the marks can be removed if the text is edited or copied in certain ways.

“We know that no watermarking system is perfect, but we believe that textGrain is an important step towards greater transparency,” the company said.

The launch comes at a time when the EU is preparing to implement the Artificial Intelligence Act, which requires companies to label AI-generated content.

OpenAI said it is working with regulators and other companies to develop common standards for identifying AI-generated content.

“We believe that transparency is essential for building trust in AI, and we are committed to working with others to achieve this goal,” the company said.