Anthropic details how Claude's text watermarks work

Anthropic published a blog post on Friday, August 15, 2026, addressing basic questions regarding how it will watermark text generated by its AI chatbot Claude. The post covers the operational mechanics of the watermarking process, whether it can be circumvented through editing, and its specific impact on programming code. This move follows the company's announcement earlier in the week that it would implement watermarking to comply with the EU AI Act's Transparency Code, which mandates that AI companies employ systems enabling the identification of AI-generated content.

The announcement sparked heated debates among users. On Reddit, one poster characterized the decision as a conspiracy against innocent Claude users, while another claimed that opposition to the measure stems solely from a desire to deceive people. Additionally, Business Insider reported that dozens of users on X claimed they cancelled their Claude subscriptions as a direct result.

Anthropic's post begins by explaining that when making low-stakes choices—such as selecting between overcast and grey to describe weather conditions—Claude embeds a pattern in its responses that remains undetectable to human readers while being readable by anyone possessing the corresponding decoding key. The company explicitly stated that watermarking does not impact output quality, rendering watermarked responses completely indistinguishable from unwatermarked ones from a reader's perspective.

Stock photo for illustration only, not from the actual event

More specifically, Anthropic confirmed it will utilize the SynthID-Text approach outlined by the Google DeepMind team in 2024, alongside plans to release a dedicated watermark detection API. The company also emphasized that watermarking is fundamentally distinct from AI detection approaches utilized by firms like Pangram, which search for specific writing tells and structural patterns to identify AI usage.

"Light editing probably won't remove the watermark completely, while a complete rewrite where every word is replaced will."

Anthropic

Regarding whether users can rewrite text to obscure the watermark, Anthropic indicated that while possible, light editing will likely fail to eliminate the watermark entirely, whereas a complete rewrite replacing every single word would succeed. In the latter scenario, however, the company noted it becomes debatable whether the resulting text can still be classified as AI-generated. For text merely proofread or heavily edited by Claude, the detection outcome depends on the text length and modification extent, with light edits leaving virtually no room for the watermark to attach to human-authored words.

Concerning code generation, Anthropic explained that software code will carry significantly fewer watermarks than standard text because the model must generate functional code, leaving fewer arbitrary choices among valid options. Watermarks can still be utilized in areas permitting arbitrary choices, such as code comments, but will exert a negligible effect on the actual code produced. Furthermore, Anthropic noted that Claude will not operate in isolation, as other major model developers have signed the same Code of Practice and will implement their respective watermarking mechanisms.

Source: TechCrunch