Anthropic to Use Google’s Watermarking Tech for Claude
The story in brief
Anthropic announced it will use a version of Google DeepMind’s SynthID-Text system to add invisible watermarks to Claude-generated text, complying with the EU’s AI transparency rules. The watermarking method embeds detectable patterns using probabilistic choices, making the markings undetectable to readers but identifiable with a key. The EU’s AI Act requires transparency from major AI developers, impacting companies like Anthropic and Google.