Claude Adds Invisible Text Watermarks Surviving Copy-Paste for EU AI Act Compliance
Commercial-Equal2238 · reddit · 2026-08-13
Anthropic is rolling out invisible text watermarks for Claude models launched on or after August 2. The watermark is embedded directly into the text, survives copy-pasting, and may persist through light editing to comply with the EU AI Act's transparency requirements. Supported image files will also receive C2PA signed provenance metadata.
Applied at the model level, the feature covers all platforms including the API and Claude Code globally. This raises concerns for commercial app builders, as even lightly edited content (like proofreading) will carry the mark, passing directly to end-users. It remains to be seen if other major labs like OpenAI and Google will follow suit with similar text-level solutions.
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
- Post-training Yandex AliceAI-80B-A3B from scratch: a NaN bug in custom V100 kernels killed one run — jjusko20 · 2026-10-03