How Anthropic plans to watermark Claude's AI-generated text
Anthropic implements invisible watermarking for Claude AI text to comply with EU regulations.
Anthropic is implementing an invisible watermarking system for its Claude AI-generated text to comply with the EU's Code of Practice, which requires AI-generated content to be identifiable. While initially driven by EU regulations, the watermarking will be applied globally. The system, based on Google DeepMind's SynthID-Text approach, subtly alters the randomness during text generation rather than adding visible markers or hidden characters, allowing for detection by those with the correct key.
AI summary of the linked source. Not reviewed by an editor. How this works
Run Google?
Get an email when a new story names Google, usually within the hour.
Free. Your list stays private and never appears in a subject line. One click stops it. How Vendor Watch worksPrivacy
The lesson from this incident
Full lessonAnthropic's adoption of invisible watermarking for Claude-generated text is primarily driven by the EU's AI Act Code of Practice, which mandates that AI-generated content be identifiable to end users and regulators. The underlying risk this addresses is the potential for AI-generated misinformation, fraud, or manipulation when synthetic content cannot be distinguished from human-authored content. By leveraging a cryptographic-key-based approach (SynthID-Text), Anthropic ensures detection is controlled and tamper-resistant rather than relying on easily stripped metadata. This matters because as AI-generated content proliferates, the absence of provenance signals creates significant risks for trust, accountability, and informed consent across digital ecosystems.