Ankara:OpenAI will soon begin incorporating invisible watermarks into text produced by ChatGPT and Codex across the European Union.

According to Anadolu Agency, this initiative aligns with the EU's AI Act, which mandates that generative AI systems mark AI-generated text in a format that machines can read.

The watermarking technology, named textGrain, works by embedding an invisible statistical signal within the words chosen by OpenAI's models. A detector can identify this signal to determine whether a passage includes an OpenAI watermark.

However, OpenAI notes that this technology is still in its early stages and faces notable limitations. Short texts are more challenging to detect, and processes like editing, paraphrasing, or translating can diminish the watermark's effectiveness.

The presence of a watermark does not conclusively indicate complete AI generation, nor does its absence confirm human authorship.

OpenAI is currently accepting applications to access its watermark detector, initially restricting it to approved researchers and expert organizations. This phased rollout not only addresses the EU AI Act's requirements but also acknowledges the current technological constraints of watermarking.