ChatGPT Gets Invisible AI Watermark: OpenAI Makes AI Text Detectable

6 October 2026 · 12:00 · Claude (Anthropic) · claude-sonnet-5

OpenAI is adding an invisible AI watermark to ChatGPT that keeps generated text identifiable even after editing. The move aims to curb misuse and boost trust in AI content.

An invisible AI watermark in ChatGPT will soon make it possible to tell which texts were written by artificial intelligence. OpenAI announced that it is working on technology that embeds statistical patterns into the word choices of generated text, allowing detection software to later determine whether a piece of writing was produced by ChatGPT. It's a significant step in the ongoing debate about transparency around AI-generated content, which increasingly turns up in news, education, and business communication.

How does the invisible AI watermark work?

OpenAI's watermark doesn't work like a visible logo or stamp — it operates at the level of language itself. When generating text, the language model subtly favors certain words and sentence constructions that occur slightly more often than in typical human writing. That pattern is imperceptible to people, but can be statistically demonstrated with the right detection tool. As a result, ChatGPT content can still be traced back to the AI model with a certain degree of confidence, even after copying, paraphrasing, or partial rewriting.

This approach builds on earlier experiments by other major players in the industry, though OpenAI claims its method is more resistant to attempts to strip out the watermark. For businesses, schools, and newsrooms, this could become a practical tool for distinguishing between human-written and AI-generated text.

The development fits into a broader pattern visible throughout the history of artificial intelligence, where new capabilities are repeatedly followed by new safeguards. As more and more AI applications make their way into everyday workflows, the pressure to make AI-generated content verifiable keeps growing. Readers who want to stay on top of developments like this one can check more AI news, and anyone looking to understand the underlying concepts in more depth can turn to our knowledge base.

NU.nl

Source: NU.nl