Claude watermark plan unsettles covert AI users
Anthropic’s plan to embed invisible watermarks in text generated by future Claude models has triggered a sharp debate among heavy AI users who fear their reliance on the chatbot could become detectable in professional, academic and creative work. The company says the technology will allow authorised detection systems to estimate whether Claude was involved in producing a passage, without…
Anthropic, the company behind the Claude AI model, has announced plans to embed invisible watermarks into text generated by future versions of Claude. This decision has sparked controversy among users who fear their use of Claude could become detectable in various professional, academic, and creative contexts. The watermarks will be created based on subtle statistical patterns in Claude's word choices, rather than adding hidden characters or altering its visible appearance.
This approach is based on Google DeepMind's SynthID-Text technology and will be present from the model's launch for future models, with support for older models being added over the coming months.
The controversy stems from concerns that the watermarks could expose the use of AI assistance in various fields, potentially misleading assumptions about authorship. Critics argue that AI systems are increasingly used as productivity tools, and marking their contribution could lead to confusion about who wrote the material. Anthropic acknowledges that extensive editing may not create enough watermarked text to produce reliable detection results, especially for light edits or short passages.
However, the company maintains that the watermark will only indicate the likelihood that Claude contributed to the material, not prove that it was entirely created by the AI.
The watermark applies to words selected by Claude, meaning that minor grammar or punctuation corrections may not leave a detectable signal. It is also expected to be weaker in code, where exact syntax requires little scope for alternative word choices. However, natural-language portions of code may carry stronger signals. Anthropic emphasizes that the watermark does not contain any personal information, such as account or organization data, and does not change ownership of generated material.
The company is developing a watermark detection API to allow approved systems to examine text for the pattern, but it will not prove that Claude wrote an entire document. The watermark system is being introduced globally as AI providers prepare for transparency obligations under the EU AI Act, which calls on providers to make machine-generated material identifiable using technical marking methods.
Written by urgent.news from Arabian Post's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
