AI Tech / news
Anthropic details how Claude's text watermarking will work
Anthropic has published a blog post explaining how it will watermark text generated by its Claude chatbot, a move tied to the EU AI Act's Transparency Code. The company says the watermark is invisible to readers but detectable with a key and does not affect output quality.
Anthropic published a blog post Friday answering basic questions about how it will watermark text produced by its Claude chatbot. The post covers how the watermarking works, whether it can be hidden through editing, and its implications for code.
The company explained that Claude creates a detectable pattern when making "low-stakes choices," such as selecting between words like "overcast" and "grey" to describe weather. This pattern is imperceptible to readers but can be identified by anyone who holds the encoding key.
Anthropic said watermarking does not impact the quality of Claude's output, and that to a reader, a watermarked response is indistinguishable from an unwatermarked one.
The clarification follows the company's earlier announcement this week that it would introduce watermarking to comply with the EU AI Act's Transparency Code, which requires AI companies to use systems that make AI-generated content identifiable.
The move has sparked debate among Claude users. Reactions on Reddit and X ranged from accusations of a conspiracy against innocent users to arguments that the only reason to oppose watermarking is to deceive others. Business Insider reported that dozens of users on X claimed to have canceled their Claude subscriptions in response.