Following a report from The Wall Street Journal suggesting that OpenAI possesses a tool capable of accurately identifying essays written by ChatGPT, the company has provided some insight into its research on text watermarking and explained why it hasn’t released the detection method. According to the report, internal debate has prevented the release of the tool, even though it is reportedly “ready.” In an update to a May blog post published on Sunday and noticed by TechCrunch, OpenAI stated, “Our teams have developed a text watermarking method that we continue to consider as we research alternatives.”
OpenAI explained that watermarking is just one of several solutions it has explored, including classifiers and metadata, as part of “extensive research on the area of text provenance.” While the watermarking method has shown high accuracy in some scenarios, it struggles with certain tampering techniques, such as using translation systems, rewording with another generative model, or inserting and then removing special characters between words. Additionally, OpenAI noted that text watermarking could “disproportionately impact some groups,” potentially stigmatizing the use of AI as a writing tool for non-native English speakers.
The company emphasized that it is carefully considering these risks. OpenAI also mentioned that it has prioritized developing authentication tools for audiovisual content. An OpenAI spokesperson told TechCrunch that the company is taking a “deliberate approach” to text provenance due to “the complexities involved and its likely impact on the broader ecosystem beyond OpenAI.”
We have helped 20+ companies in industries like Finance, Transportation, Health, Tourism, Events, Education, Sports.
