Anthropic to Add Invisible Watermarks to Claude Text for EU Compliance
Summarized by AI from reporting by The Verge AI, published under our editorial policy.
Anthropic will use invisible watermarks based on Google DeepMind's SynthID-Text to comply with EU AI transparency rules. These watermarks embed detectable patterns in Claude's text using word probabilities, enabling identification of AI-generated content without affecting readability.
Key takeaways
- Anthropic will use invisible watermarks based on Google DeepMind's SynthID-Text to comply with EU AI transparency rules.
- The watermarking system embeds detectable patterns in Claude's text by manipulating word probabilities.
- Watermarks will be detectable using specialized tools but invisible to human users, preserving text readability.
Anthropic has announced that it will add invisible watermarks to text generated by its AI model, Claude, to comply with Europe's AI transparency rules. This system, a version of the SynthID-Text approach, will create detectable patterns using wording probabilities, ensuring that AI-generated text can be identified.
How the SynthID-Text Watermarking System Works
Anthropic's watermarking system is based on the SynthID-Text approach, an open-source technology developed by Google DeepMind. This method embeds detectable patterns in the text by manipulating the probabilities of certain words and phrases. The watermarks are invisible to the human eye but can be detected using specialized tools, making it possible to identify AI-generated content.
Compliance with EU AI Transparency Regulations
The European Union has implemented strict AI transparency rules, requiring companies to disclose when AI-generated content is used. By adding watermarks to Claude's output, Anthropic aims to comply with these regulations and ensure that users and platforms can identify AI-generated text. This move is part of a broader effort to enhance transparency and accountability in the AI industry.
Impact on Users and Platforms
For everyday users, these watermarks will not be visible, so the readability and usability of Claude's text will remain unchanged. However, platforms and content moderators will be able to detect these watermarks, helping them identify AI-generated content and ensure compliance with regulations. This could also help prevent the spread of misinformation and deepfake content.
What You Can Do Today
If you use Claude or any other AI text generation tools, you can check for updates from Anthropic or the platform you're using to see if they have implemented watermarking technology. You can also stay informed about EU AI regulations and how they might affect the tools and services you use. For more details, visit Anthropic's official website or the EU's AI regulation page.
Frequently asked
- Will the watermarks affect the readability of Claude's text?
- No, the watermarks will be invisible to human users and will not affect the readability of the text.
- How will platforms detect these watermarks?
- Platforms and content moderators will use specialized tools to detect the watermarks in AI-generated text.
- Are there any other AI models using similar watermarking technology?
- The source story does not mention other AI models using similar watermarking technology, but SynthID-Text is an open-source approach developed by Google DeepMind.