Anthropic is taking another step toward more transparent AI by explaining how it plans to watermark text generated by Claude. The company says the system will use a version of SynthID-Text, an open-source watermarking approach developed by Google DeepMind.
Unlike visible labels, this technique works by introducing subtle, detectable patterns in word choices and probabilities. That means AI-generated text can carry a machine-readable signal while remaining natural for people to read.
Why it matters
The feature is being introduced as part of Anthropic’s efforts to comply with the European Union’s AI Act, which requires synthetic text, images, audio, and video to include machine-readable marks. Anthropic is also adding C2PA support for Claude-processed images, strengthening provenance signals across multiple content types.
- Invisible text watermarks can help platforms and auditors identify AI-generated material.
- Open-source foundations like SynthID-Text make transparency tools easier to evaluate and improve.
- Compliance-driven deployment turns AI safety and provenance research into real-world product features.
While watermarking is not a complete solution to misinformation or attribution challenges, Anthropic’s rollout is a meaningful win for responsible AI. It shows major AI developers moving from policy promises to practical systems that help people better understand when content may be machine-generated.