Anthropic Adds Text Watermarking to Claude Models Worldwide

Anthropic has started embedding imperceptible, machine-readable watermarks into text generated by its newest Claude models, applying the change globally rather than only in the EU. The trigger is Article 50 of the EU AI Act, which became enforceable on August 2, 2026. For marketing teams producing AI-assisted content at location-level volume, content provenance just moved from a theoretical concern to a technical fact.

What Anthropic did

Anthropic is using the SynthID-Text approach that Google DeepMind published in 2024, which adjusts token selection during generation in a way that is statistically detectable but invisible to readers. The watermarking covers new Claude models across Claude, the Claude API, Claude Code, Claude Cowork and Claude Tag. The company plans to release a detection API and is working to retrofit models released before August 2 during the AI Act transition period.

The regulatory driver is Article 50 of the EU AI Act, which requires providers of generative AI systems to mark outputs in machine-readable formats. Anthropic could have geofenced the feature to Europe. It chose not to, which means the marker is present regardless of where your team sits.

Anthropic has been careful about what the watermark does and does not prove. It cannot confirm authorship and it cannot rule out AI involvement. Reporting on the rollout has surfaced two details that matter more operationally than the policy itself: the marks survive copy and paste, and they can appear even when Claude only corrected spelling.

What this Means

This is not a search penalty, and nothing in the announcement suggests Google or Meta will start downranking watermarked text. Google’s own guidance has consistently focused on content quality rather than production method. Treat the risk as reputational and contractual before treating it as an SEO problem.

The realistic near-term exposure sits in three places: client or franchisee contracts with AI disclosure clauses, regulated verticals where advertising claims must be attributable to a human reviewer, and public claims your brand has already made about how its content is produced.