Anthropic published a blog post Friday to explain how it will watermark text generated by its chatbot Claude. The company addressed basic questions about the mechanics, whether editing removes the mark, and the impact on code.
Users have debated the move since the firm announced it earlier this week to comply with the EU AI Act’s Transparency Code. This regulation requires AI companies to use systems that make it possible to identify AI-generated content.
On Reddit, one poster characterised this as a conspiracy against innocent Claude users, while another claimed, “The only reason you wouldn’t want this is to lie to people.” Business Insider reports that “dozens” of users on X have claimed to cancel their Claude subscriptions as a result.
How the mark works
The new post starts with a general overview of the watermarking concept. When making “low-stakes choices” — like choosing between the words “overcast” and “grey” to describe the weather — Claude can create a pattern in its responses that is “undetectable to the reader, but is detectable to anyone who has a key that encodes it.”
“Watermarking does not impact the quality of Claude’s output,” the company said. “To a reader, a watermarked response is indistinguishable from an unwatermarked one.”
More specifically, Anthropic said it will be using the SynthID-Text approach that the Google DeepMind team outlined in 2024. It plans to release a watermark detection API. The firm noted that watermarking is distinct from the AI detection approaches offered by companies like Pangram that look for “tells” in the writing (like the construction “his isn’t [X], it’s [Y]”) to reveal AI usage. “Picking up on these patterns is fundamentally different from checking for a watermark.”
Can you hide it?
Anthropic said it is possible to rewrite the text to hide the watermark. However, “light editing probably won’t remove the watermark completely,” while “a complete rewrite where every word is replaced will.”
“In the latter case, of course, it’s arguable whether the text can any longer be described as AI-generated,” the company said.
Edited or proofread text
As for whether the watermark will be detectable in text that was only proofread or edited by Claude, Anthropic said that will depend on “the length of the text and how heavily Claude has edited it.” If it is only been lightly edited, “nearly all the words” will have been written by the human author and “there’s very little (if anything) for the watermark to attach to.”
Code generation
Code should have less of a watermark than other text because the model will need to create working code and will not have the freedom to choose between a variety of equally valid options.
“Having said that, in areas where there is an arbitrary choice between particular words or terms within the code, the watermark can be used, such as comments within code,” Anthropic said. “But by definition, it will have a negligible effect on the actual code produced.”
Other providers
Anthropic said that Claude will not be the only AI chatbot to generate watermarked text. “Other major model developers have signed the same Code of Practice and will be implementing their own watermarks.”
