Updated August 11, 2026
The verified list is shorter than most search results suggest. These statuses use public provider documentation and distinguish text watermarks from image, audio, and video watermarks.
Scan copied textConfirmed by provider
Supported Claude models launched on or after August 2, 2026
Anthropic says embedded text watermarks apply across Claude, Claude Platform API, Claude Code, Claude Cowork, Claude Tag, and supported access through AWS, Google Cloud, and Microsoft Foundry. Existing older models are still being transitioned.
Read official sourceConfirmed by provider
Gemini app and web experience
Google DeepMind says SynthID adjusts token probabilities during generation. This is a statistical watermark, not hidden Unicode that can be removed with character cleanup.
Read official sourceOpenAI publicly describes text watermarking research, but its current deployed provenance documentation focuses on images and other media.
Microsoft documents watermarks for images, video, and audio. We found no official confirmation for ordinary Copilot chat text.
xAI documents Grok watermarks for generated images and videos, not ordinary chat text.
Meta documents invisible watermarks for generated images, not a deployed text watermark for ordinary Meta AI or Llama responses.
No official public confirmation of a deployed watermark in ordinary generated text was located as of August 11, 2026.
A local Unicode scan cannot detect Claude or Gemini statistical watermarks. It can find literal invisible characters and safely clean supported formatting artifacts. A character finding is not proof of who authored the text.