Anthropic: AI-written text will now be easily detected, with a special watermark hidden within Cloud content.
To make it easier to distinguish between human- and AI-written content and to comply with the European Union AI Act, AI company Anthropic is adding a watermark to the text of its Claude chatbot. But this watermark isn't a visible label or code.
Until now, you might have seen watermarks on photos and videos created by AI. But now, Anthropic, the company behind the AI chatbot "Claude," is adding a watermark to text written by its AI model. This is being done to make it easier to identify whether Cloud was used to create a text, because it is becoming increasingly difficult to differentiate between AI-created content and human-written content. But this watermark, like a symbol or code, will not be visible to everyone separately on the text.
The watermark will not be clearly visible on the content
Anthropic says this watermark is being implemented worldwide to comply with the European Union's AI Act. This act requires AI-created content to be marked in a way that is machine-readable. Anthropic said it is implementing this system worldwide.
However, it should be noted that the watermark on the text will not be a visible label or hidden character. Cloud AI writes its answers by selecting words one by one. When it has multiple correct word options, it uses a special system to select words. This word selection becomes its identity, the watermark. Anthropic said that these selections leave a pattern that is invisible to the reader, but can be recognized by someone with the correct key.
Readers won't know the difference.
The company stated that the watermark does not significantly impact the quality or content of Cloud's output. It also stated that readers will not be able to tell the difference between watermarked and unwatermarked text. No additional characters or tokens are added to the response, meaning the watermark does not significantly impact Cloud's performance or speed.
copy-pasted the text, and may not be removed even with minor editing. However, it's not impossible to remove. Anthropic says that if the entire text is rewritten and every word is changed, the watermark will disappear. The company also explains that the watermark only indicates that the cloud had some level of involvement with the content. This means that the watermark does not prove that the cloud wrote the entire article, essay, or document.
