Anthropic published a blog post on August 15, 2026, explaining how watermarking will work for text generated by its Claude chatbot, addressing user questions about detection, editing, and the impact on code.
The company said it will use the SynthID-Text approach developed by Google DeepMind in 2024. The system works by exploiting low-stakes word choices — for example, selecting “overcast” over “grey” — to embed an invisible pattern in Claude’s responses. “To a reader, a watermarked response is indistinguishable from an unwatermarked one,” Anthropic said, adding that watermarking does not affect output quality. Anthropic also plans to release a watermark detection API.
The move is driven by compliance with the EU AI Act’s Transparency Code, which requires AI companies to implement systems capable of identifying AI-generated content. Anthropic noted that other major model developers have signed the same Code of Practice and will implement their own watermarks.
The announcement has drawn mixed reactions. On Reddit, some users characterized the watermarking as an overreach, while others argued it exists to prevent deception. Business Insider reported that dozens of users on X claimed to have canceled their Claude subscriptions in response.
On the question of whether editing can remove the watermark, Anthropic said light edits probably won’t eliminate it entirely, but a complete rewrite replacing every word will. “In the latter case, of course, it’s arguable whether the text can any longer be described as AI-generated,” the company said. For text only lightly edited by Claude, the watermark may have little to attach to, depending on the length and extent of Claude’s involvement.
Code generated by Claude will carry less of a watermark than prose, because producing functional code leaves the model fewer arbitrary word choices. Anthropic said the watermark could still appear in code comments, but will have a “negligible effect on the actual code produced.”
Source: TechCrunch