Decision to add secret signals to ChatGPT text
Published 1 day ago | By Nouman Shakeel
OpenAI has announced the introduction of a new technology designed to make text generated by ChatGPT identifiable. Under this system, AI-generated writing will contain subtle, hidden signals—imperceptible to the average reader but detectable by automated systems.
According to the company, this technology—dubbed "textGrain"—will introduce minor statistical variations into ChatGPT’s word-selection process. The aim is to embed a covert signal that allows the text to be identified later as AI-generated.
Imperceptible to the Average Reader
OpenAI states that this watermarking method will create subtle differences in the text that typically go unnoticed by readers. However, automated detection systems will be able to identify AI-generated content by spotting these specific signals.
This development comes as Anthropic has also announced plans to incorporate similar technology into its AI model, Claude.
These moves by both companies are being linked to the European Union’s AI Act, which mandates that developers of generative AI products make their content identifiable via automated systems.
User Objections to the Technology
OpenAI’s decision has faced objections from some users. Critics argue that such a system may not prove entirely reliable and could result in unnecessary interference with the AI's text-generation process. For now, OpenAI has decided to introduce this watermarking technology only in European Union countries, whereas the scope of Anthropic's initiative differs.
