tech
Anthropic's text watermarks signal new front in AI detection
Claude-generated text and files will carry watermarks to comply with new European Union transparency regulations.
TL;DR
- Anthropic is adding machine-readable watermarks to Claude-generated text and files to comply with EU transparency regulations.
- Text watermarks are embedded patterns within the text, claimed to be imperceptible.
- File metadata will include digital signatures confirming Claude's processing.
- Limitations include potential false positives for AI-assisted content and non-detectable watermarks in heavily rewritten or short texts.
- These changes are in response to Article 50 of the EU AI Act, mandating AI disclosures.
- Other companies like OpenAI are also adapting their compliance approaches, with a focus on images and audio.
- Digital platforms like LinkedIn, Substack, and Snap are also accelerating their AI identification measures.
- AI providers have until December 2 to bring legacy models into compliance with EU rules.