Anthropic watermarks Claude using Google's SynthID-Text
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic clarified its text watermarking system for Claude uses "a version of the SynthID-Text approach," an open-source watermarking technology developed by Google DeepMind that creates detectable patterns using wording probabilities.
- The watermarking works by steering low-stakes word choices (e.g., selecting between "overcast" or "grey" to follow "The weather today was cold and…") through a randomness source that combines a key with preceding words — producing a pattern undetectable to readers but identifiable to key-holders.
- Anthropic said the watermarks will not raise Claude's cost for users and will have no practical impact on the quality or content of Claude's outputs.
- The watermarking feature, alongside C2PA support for Claude-processed images, is being introduced to satisfy the EU AI Act's requirement that synthetic audio, image, video, and text include machine-readable marks identifying content as AI-generated.
- Google's Gemini chatbot has supported SynthID-Text since 2024, and OpenAI — which has not detailed text watermarking plans in its AI Act compliance roadmap — will also be subject to the EU law's requirements.
- Anthropic noted that under the EU's AI transparency rules, Claude will not be the only model introducing text watermarks; the requirements affect all major AI developers.
Why it matters: The EU AI Act makes watermarking mandatory for any major AI developer serving European users, forcing synthetic text and images to carry machine-readable identification. By basing Claude's system on Google's open-source SynthID-Text — the same foundation Gemini has used since 2024 — Anthropic is taking the cheapest compliance path while making AI-generated text traceable to key-holders under EU rules yet indistinguishable to human readers.
Ask SkimNews
