Anthropic has released more information about its planned watermarking for Claude, addressing how the system works, whether it can be edited away, and its effect on code.
Anthropic has shared additional details about how it intends to watermark text written by its AI assistant Claude. In a blog post published on Friday, the company addressed a series of frequently asked questions surrounding the technology.
The clarification comes days after Anthropic first announced that it would introduce watermarking to meet the requirements of the European Union’s AI Act. Specifically, the EU AI Act’s Transparency Code obliges AI providers to put systems in place that make AI generated content identifiable.
According to the blog post, the company sought to explain how the watermarking would function, whether it could be stripped out through editing, and what the system would mean for code produced with Claude.
Since the initial announcement earlier this week, Claude users have been actively debating the move, with opinions divided over its potential effectiveness and impact.
The new blog post forms part of Anthropic’s public engagement as it prepares to align Claude with the EU’s transparency requirements.