Anthropic has detailed its implementation of invisible watermarks for text generated by its AI model, Claude, to align with the European Union's AI transparency regulations. This system utilizes a modified version of Google's open-source SynthID-Text technology, which embeds detectable patterns in the text based on word probabilities without affecting the output's quality or cost for users.
The watermarking process relies on low-stakes word choices to create a unique pattern that remains undetectable to readers but can be identified by those with the appropriate key. This initiative reflects a broader trend among AI developers, as compliance with the EU's AI Act is expected to influence other models, including Google's Gemini and potentially OpenAI's ChatGPT.
Watch for how other AI developers respond to the EU's transparency regulations, particularly Google's Gemini and OpenAI's ChatGPT. Anthropic's watermarking could set a precedent, influencing user trust and content verification in AI-generated text across platforms.