\n<\/p>\n
Anthropic published a blog post<\/a> Friday seeking to answer some basic questions about how it will watermark the text generated by its chatbot Claude. Such as: How will the watermarking actually work? Can it be hidden with editing? And how does this affect code?<\/p>\n Claude users have been debating the move since the company revealed earlier this week<\/a> that it would be doing this watermarking to comply with the EU AI Act\u2019s Transparency Code, which requires AI companies to use systems that make it possible to identify AI-generated content.<\/p>\n On Reddit, for example, one poster characterized this as a conspiracy against innocent Claude users<\/a>, while another claimed, \u201cThe only reason you wouldn\u2019t want this is to lie to people.\u201d And Business Insider reports<\/a> that \u201cdozens\u201d of users on X have claimed to cancel their Claude subscriptions as a result.<\/p>\n Anthropic\u2019s new post starts with a general overview of the watermarking concept, explaining that when making \u201clow-stakes choices\u201d \u2014 like choosing between the words \u201covercast\u201d and \u201cgrey\u201d to describe the weather \u2014 Claude can create a pattern in its responses that is \u201cundetectable to the reader, but is<\/em> detectable to anyone who has a key that encodes it.\u201d<\/p>\n \u201cWatermarking does not impact the quality of Claude\u2019s output,\u201d the company said. \u201cTo a reader, a watermarked response is indistinguishable from an unwatermarked one.\u201d<\/p>\n More specifically, Anthropic said it will be using the SynthID-Text approach that the Google DeepMind team outlined in 2024<\/a>, and that it plans to release a watermark detection API. It also noted that watermarking is distinct from the AI detection approaches offered by companies like Pangram<\/a> that look for \u201ctells\u201d in the writing (like the construction \u201chis isn\u2019t [X], it\u2019s [Y]\u201d) to reveal AI usage: \u201cPicking up on these patterns is fundamentally different from checking for a watermark.\u201d<\/p>\n Could someone just rewrite the text to hide the watermark? Anthropic said it\u2019s possible, but \u201clight editing probably won\u2019t remove the watermark completely,\u201d while \u201ca complete rewrite where every word is replaced will.\u201d<\/p>\n \u201cIn the latter case, of course, it\u2019s arguable whether the text can any longer be described as AI-generated,\u201d the company said.<\/p>\n As for whether the watermark will be detectable in text that was only proofread or edited by Claude, Anthropic said that will depend on \u201cthe length of the text and how heavily Claude has edited it.\u201d If it\u2019s only been lightly edited, \u201cnearly all the words\u201d will have been written by the human author and \u201cthere\u2019s very little (if anything) for the watermark to attach to.\u201d<\/p>\n Code, meanwhile, should have less of a watermark than other text, because the model will need to create working code and won\u2019t have the freedom to choose between a variety of equally valid options.\u00a0<\/p>\n \u201cHaving 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,\u201d Anthropic said. \u201cBut by definition, it will have a negligible effect on the actual code produced.\u201d<\/p>\n Anthropic also said that Claude won\u2019t be the only AI chatbot to generate watermarked text, as \u201cother major model developers have signed the same Code of Practice and will be implementing their own watermarks.\u201d<\/p>\n<\/div>\n When you purchase through links in our articles, we may earn a small commission<\/a>. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n