Last reviewed: August 17, 2026. Rollout and detector availability may change; source links below are the authority.
Claude Text Watermark: 2026 Rollout and Detection Status
Timeline
| Date | Event | Source |
|---|
| July 2026 | Anthropic signed the EU Code of Practice on Transparency of AI-Generated Content. | Anthropic |
| August 2, 2026 | Claude models launched in the EU from this date support machine-readable marking at launch; Anthropic applies supported marks worldwide. | Claude Help Center |
| August 12, 2026 | Independent coverage explained the new-model rollout and the difference between statistical text marks and file provenance metadata. | TechRadar |
| August 14, 2026 | Anthropic identified its method as a version of SynthID-Text and explained how keyed word-choice patterns work. | Anthropic |
Current Published Status
| Aspect | Earlier models or position | Current published position |
|---|
| Model rollout | Models launched before August 2, 2026 fall under a transition period. | New supported models carry marks at launch; Anthropic is adding support to older models. |
| Text mechanism | No implementation detail was cited on this site. | Anthropic identifies a version of SynthID-Text that changes keyed word-choice randomness and adds no hidden characters. |
| Detection access | No public Anthropic detector was available. | A detection API has been announced but was not public as of August 17, 2026. |
| Meaning of a result | Often discussed as proof of AI authorship. | Anthropic says a mark only suggests Claude may have processed the content. |
- Invisible Unicode characters are not Claude's official text watermark.
- Short, factual, code-heavy, or lightly proofread passages may contain too little signal for reliable detection.
- Heavy rewriting can weaken the signal, but this site does not provide watermark removal or bypass tools.
- A detected mark cannot establish authorship, ownership, or whether Claude wrote rather than edited the text.
Related Claude Text Watermark Guides