OpenAI confirms exploring text watermarking for ChatGPT

Despite the tool being reportedly ready, OpenAI has not released it due to internal debate over potential impacts.

In a recent update to a blog post, OpenAI detailed their exploration of watermarking, among other methods like classifiers and metadata, to verify text origins. While the watermarking technique has shown high accuracy in certain scenarios, it struggles with tampering methods "like using translation systems, rewording with another generative model, or asking the model to insert a special character in between every word and then deleting that character."

OpenAI also expressed concerns about the potential negative effects of watermarking, particularly its disproportionate impact on non-native English speakers using AI for writing assistance. The company emphasized the importance of carefully considering these risks and the broader implications for the ecosystem.

OpenAI has prioritized developing authentication tools for audiovisual content and is taking a "deliberate approach" to text provenance, according to a statement to TechCrunch.