Anthropic adopts Google's SynthID-Text to watermark Claude outputs

Anthropic has clarified how it plans to watermark Claude-generated text to comply with Europe's AI transparency rules. On Friday, the company revealed that Claude's text marking system is “a version of the SynthID-Text approach”—an open-source watermarking technology from Google DeepMind that creates detectable patterns using word-choice probabilities. The feature, along with C2PA support for Claude-processed images, is designed to meet obligations under the EU AI Act, which requires synthetic audio, image, video, and text to carry machine-readable marks indicating they're artificially generated or manipulated.

Anthropic says the watermarks won't make Claude more expensive or have “any practical impact on the quality or content of Claude's outputs.” The mechanism works by exploiting low-stakes word choices—for example, after “The weather today was cold and…,” the next word is unlikely to be “sugary” but could reasonably be “overcast” or “grey.” Normally, such choices are settled by a random number generator. With watermarking, the randomness is instead derived from a secret key plus preceding words, leaving an undetectable pattern that anyone with the key can verify. Anthropic notes that other major AI developers are also affected: Google's Gemini has supported SynthID-Text since 2024, and OpenAI, though not yet detailing text watermarking plans, will also need to comply with the EU's requirements.