Anthropic Rolls Out Invisible Text Watermarks for Claude Globally
AGI Hunt · wechat · 2026-08-11
Anthropic has begun embedding invisible watermarks in all Claude text outputs to comply with the EU AI Act, with the mechanism active globally.
The watermarking operates on two levels:
- Text Watermark: Embedded directly into the text during generation by tweaking vocabulary choices. It survives copy-pasting and light editing but may degrade with heavy modification or very short texts.
- File Metadata: Appends C2PA-compliant signatures to generated files (SVG, PNG, etc.), which are lost upon screenshotting or format conversion.
This feature covers all products and cloud platforms. Because many users leverage Claude for editing or translation, the presence of a watermark only indicates potential processing by Claude, not absolute authorship. For comparison, Google Gemini uses a similar token-biasing technique, while OpenAI previously abandoned its watermarking plans over concerns about text quality and user churn.
More from Models
- First Model to Pass the Video Turing Test: Half of Users Thought It Was Human — Puzzleheaded-King584 · 2026-10-03
- Using System One models in Swift: fast, deterministic decisions via Apple Foundation Models — rxwei · 2026-10-03
- Linux Kernel CVEs Surge From ~500 to 1500+ Per Release, LLMs Blamed for Bulk of the Rise — burny_tech · 2026-10-03
- Developer Complains OpenAI's Coding Model Endlessly Scopes Creeps Instead of Finishing Tasks — DavidWells · 2026-10-03
- Sonnet 5 Spotted in Google Antigravity Backend, Which Still Runs Sonnet 4.6 — brandon_galang · 2026-10-03
- Every's Dev Day chat with Matthew Berman: budget gone the moment he tried Ultrafast — every · 2026-10-03