Anthropic's text watermarks signal new front in AI detection
Summary
Anthropic is adding hidden digital watermarks to text and media created by its AI model Claude to meet new European Union rules. These watermarks help identify AI-generated content but have some limits, such as being less effective if the text is heavily changed.Key Facts
- Anthropic’s AI model Claude now includes machine-readable “watermarks” in text and files created for users in the EU.
- The watermarks are patterns in the text that people cannot see but machines can detect.
- Media files created by Claude carry digital signatures to show they were processed by the AI.
- These measures comply with Article 50 of the EU AI Act, which requires clear disclosure of AI-generated or altered content.
- The watermarks can sometimes appear even if Claude only helped to proofread or translate human-written text.
- Watermarks may not be detectable if the AI-generated text is heavily rewritten, mixed with other texts, or very short.
- Other companies like OpenAI are also working on AI content disclosure but focus more on images and audio so far.
- Digital platforms like LinkedIn, Substack, and Snap are testing or using AI detection tools to identify AI-generated content.
- AI providers must update older AI models to meet EU rules by December 2, 2024.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.
