Updated:12:03 AM CEST Aug,12
(new)
(c) 1998-2026 Gameguru Mania
Privacy Policy statement
|
Claude Now Hides Invisible Watermarks in Every Text It Generates - briefly|
| (hx) 12:03 AM CEST - Aug,12 2026 | 
Anthropic has started embedding an invisible watermark into every piece of text produced by its Claude AI models. The company confirmed the change in an official support page after signing the EU AI Act’s transparency code for AI-generated content.
The text watermark is woven directly into the words themselves at the model level. You cannot see it, and it does not change the meaning, quality, or readability of the response. Because it lives inside the text, the mark travels when you copy and paste the content and can survive some editing.
Claude also adds digitally signed provenance metadata to certain generated files such as SVG, PNG, and JPG images. This follows the open C2PA standard and acts like a tamper-evident digital certificate that shows the file was processed by Claude and whether it was later altered.
New Claude models launched on or after August 2, 2026 carry these marks from day one. Older models are receiving the feature during a transition period. The marks apply worldwide across the Claude app, API, Claude Code, Claude Cowork, Claude Tag, and supported cloud platforms including AWS, Google Cloud, and Microsoft Foundry.
Think of the text watermark as a secret pattern hidden in the way the AI chooses and arranges words. Humans read normal sentences; computers can later check for Claude’s signature. The file metadata works more like a sealed stamp that stays attached to the image until someone strips it through conversion or screenshots.
A detected mark only signals that Claude processed the content. It does not prove Claude was the original author, because people often use the model to edit, translate, or improve existing text. Heavy rewriting, very short passages, or older unmarked models can remove or avoid the signal. Anthropic says detection tools and full technical details will be released soon.
There is no mentioned way for users to turn the marking off.
|
|
last 10 comments: All comments
|
|
)
Add your comment (free registration required)