Anthropic to Embed Invisible Watermarks in Claude Outputs for EU AI Act Compliance
dr_cintas · x · 2026-08-11
Anthropic has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content. To meet these commitments, the company plans to add machine-readable marks to Claude's outputs.
The marking mechanism includes two main components:
- Text Watermarking: An imperceptible watermark woven into the generated text without altering its meaning, quality, or readability.
- File Provenance Metadata: Generated files like .svg, .png, and .jpg will include digitally signed provenance metadata based on the C2PA open standard.
These marks will apply across all usage surfaces, including the API, Claude web interface, and Claude Code. New models launching in the EU on or after August 2, 2026, will support this at launch, while existing models are currently in a transition period for adding support. Anthropic also commits to helping third parties detect these marks.
More from Models
- Hugging Face model update: better multi-character support — OdinLovis · 2026-08-11
- Exploring Knowledge Cutoffs and Pre-training Timelines for Claude & GPT — sshh12 · 2026-08-11
- Grok 4.6 rolling out in Cursor now — daniel_mac8 · 2026-08-11
- Renaming PDF to 'final_draft' Boosts LLM Review Scores: Evaluation Weirdness — CShorten30 · 2026-08-11
- User Tests New Grok Build: Code 'Laziness' Fixed in Minecraft House Task — Angaisb_ · 2026-08-11
- Hands-on with Grok in Cursor: Fast and Agentic, but Lacks Creativity — andrew_n_carr · 2026-08-11