Confirmed
Anthropic plans to offer an API that checks whether Claude likely processed a piece of text.
Official release tracker
Claude watermark detection API
Anthropic confirmed on August 14, 2026 that a detection API is coming soon. The announcement is official. The technical specification and API access are not available yet.
Still pending
Endpoint documentation, access rules, keys, thresholds, minimum lengths, test vectors, rate limits, and pricing.
Our readiness
Our versioned provider adapter is prepared. It will remain off until official results can be reproduced and labeled accurately.
What the result will and will not mean
Anthropic says its watermark estimates whether Claude was likely involved with content. A positive result will not necessarily prove Claude wrote the original text. A negative result will not prove human authorship. Short text, editing, translation, and other transformations may affect detection.
What works today
AI Text Watermark Remover can report and clean supported literal hidden characters and formatting artifacts locally. That is useful, but it is not the same as Claude's future model-level verifier. The two results will remain clearly separated.
Open the local scanner