Search

Anthropic injects the imperceptible watermark directly into Claude's generated text at the model level, so the watermark is part of the text itself and will be present across any Claude product or surface. Because it is embedded in the text, the watermark can travel when the text is copied and pasted and may persist through some light editing. Anthropic is also working to enable detection tools that will mark positively detected text as having been processed by Claude, though such detections will not be fully conclusive. Answered

Anthropic is embedding imperceptible, model-level watermarks directly into Claude-generated text and says it is working to enable Claude users and third parties to detect those embedded watermarks, with details on detection to come. A positive detection will mark text as being "processed by Claude," but Anthropic cautions that such a result will not be fully conclusive because use of Claude for proofreading or edits does not prove original authorship. Answered

Watermarking will be applied at the model level, so it will be present no matter which Claude product or surface the text comes from. Answered

No. Many of the exposed share links have been removed from Google Search, and attempting the "site:claude.ai/share" query now produces zero results, indicating the issue appears to have been resolved. Answered