OpenAI is implementing invisible watermarks in ChatGPT — who will be affected
Kyiv • UNN
OpenAI is launching invisible textGrain watermarks initially for users in the EU. The company acknowledges that they do not prove the authorship or accuracy of the text.
The invisible, machine-readable textGrain watermark is being implemented in the output text of ChatGPT and Codex by OpenAI. But initially, only for users in the European Union. UNN writes about this, citing The Verge.
Details
The company says that such text watermarks will not become a "global standard" for now.
OpenAI states that its textGrain watermarks "match or surpass" other approaches, such as Google DeepMind's SynthID for text, which also underlies the watermarks announced by Anthropic in August. Like OpenAI, Anthropic took this step to comply with the requirements of the EU AI Act. However, not everyone was happy to learn about the new addition, the publication writes.
OpenAI also cited the results of AI tests demonstrating similar performance for both watermarked and non-watermarked text. However, the document notes that textGrain "does not guarantee reliable detection" and says that text watermarks do not confirm accuracy, identify the owner of the text, measure human contribution, or prove authorship.
OpenAI canceled the launch of a new AI model due to safety concerns29.09.26, 09:47 • 4259 views