How Claude's text watermarking works \ Anthropic
Anthropic is implementing a text watermark in future Claude models to comply with the EU AI Act. This system allows for the detection of AI-generated content without affecting output quality or cost.
Why it matters
This allows users to verify if text was likely generated by Claude via an upcoming API, increasing transparency regarding AI-authored content.
The details
- Watermarking does not identify individual users, organizations, or specific chats.
- Images will use C2PA metadata credentials instead of traditional watermarks.
- Detection is less effective on short samples, factual passages, or light edits.
Show entities and relationshipsHide entities and relationships
In this article
Key connections
Claude uses Text Watermarking
Future Claude models will incorporate text watermarking to mark AI-generated content.
Claude uses SynthID-Text
Claude's text watermark is a version of the SynthID-Text technique.
Claude attaches C2PA content credentials to generated images and files.
Claude uses Content Credentials
Claude attaches cryptographically signed content credentials to supported file types.
Claude is built with Large Language Models
Claude is a large language model that generates text word-by-word.
Google DeepMind uses SynthID-Text
Google DeepMind developed and published the SynthID-Text watermarking technique.
Show 5 more connectionsShow fewer connections
Gemini uses SynthID-Text
Google DeepMind served a model using SynthID-Text watermarking on Gemini traffic.
European Union regulates Anthropic
The EU AI Act and Code of Practice require Anthropic and other AI providers to mark AI-generated content.
SynthID-Text is related to Text Watermarking
SynthID-Text is a watermarking technique designed for text generated by AI models.
Content Credentials is built with C2PA
Content credentials adhere to the open C2PA industry standard.
Claude and Gemini are competing frontier multimodal AI model platforms.
Related events
Anthropic Implements Text Watermarking in Future Claude Models
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.