American tech firm Anthropic has revealed how its Claude artificial intelligence will incorporate invisible text watermarks to comply with Europe’s stringent AI regulations. This watermarking system is based on the SynthID-Text approach developed by Google DeepMind, which subtly alters word choices in generated text.

The technique works by making slight, undetectable changes to the probability of certain words appearing next to each other within a sentence. For example, if you were to type “The weather today was cold and…”, Claude’s choice between “overcast” or “grey” might be influenced by an embedded watermark rather than its usual randomness.

While Anthropic assures users that this won’t affect the quality of Claude’s outputs, the move mirrors similar measures from other major AI developers. Google’s Gemini chatbot has already adopted a similar system since 2024, and it is highly probable that OpenAI will follow suit to meet EU requirements.

Anthropic highlights that these watermarks are part of broader efforts to make AI-generated content more transparent and distinguishable from human-created material. However, the implementation raises questions about how much intervention in the creative process is necessary—and whether such measures are truly beneficial for users or just another layer of complexity.