Anthropic details Claude watermark amid user backlash

Anthropic announced it is implementing a text watermarking system for its Claude AI chatbot, a measure designed to comply with the European Union's AI Act. The company detailed the initiative in a blog post published Friday, August 14, following an earlier revelation earlier in the week about its plans. The move has prompted debate among users over potential impacts on output quality and privacy.

The watermarking technology works by creating an invisible pattern in the way Claude selects words when generating text. Anthropic explained that in many sentences, there are multiple acceptable word choices. The system alters the source of randomness during these low-stakes decisions, embedding a detectable signature without adding visible markers or hidden characters. "Nothing is added to the text and there are no hidden characters," the company stated. The approach is based on Google DeepMind's SynthID-Text method, and Anthropic plans to release a corresponding detection API. According to the company, internal testing found no impact on the content, creativity, or readability of Claude's output, and watermarking does not increase user costs.

However, Anthropic acknowledged the system...