Anthropic to watermark all Claude-generated text and files to meet EU rules

Anthropic's move ties directly to the EU AI Act's transparency mandates, which require AI-generated content to be identifiable. Rather than limit the change to European users, Anthropic is applying watermarking globally across its entire product surface — the API, the Claude apps, Claude Code, Claude Cowork and Claude Tag — for any model version released on or after August 2, 2026, with pre-August models to be brought into compliance by December 2, 2026.

The watermarks are machine-readable markers embedded in generated text and files, intended to let downstream systems verify provenance. Anthropic has said it will publish guidance on how customers and third parties can detect and verify the marks, though at announcement time the verification mechanism had not been detailed — a gap that fueled uncertainty.

The reaction from paying subscribers was sharp and immediate. Writers and developers worried that lightly edited, AI-assisted prose would be indelibly flagged, comparing the watermark to a 'scarlet letter,' and some raised concerns that the watermarking process itself could subtly degrade output quality. The debate cuts to a broader tension: regulators want provenance and disclosure, while professional users want AI assistance that blends invisibly into their own work.

Competitively, Anthropic is moving ahead of most rivals on visible EU compliance, which could become a selling point for enterprise and government customers who need auditable provenance — but it also risks pushing quality-sensitive individual users toward competitors that have not yet imposed comparable marking. Watch for Anthropic's verification spec and whether OpenAI, Google, and others follow with their own watermarking regimes.