Anthropic Watermarks Everything Claude Generates Worldwide to Comply With the EU AI Act

My take: Since August 2, 2026, everything Claude generates carries an invisible watermark. This isn't a voluntary move: it's compliance with Article 50 of the European Union's AI Act, which requires generative AI providers to mark synthetic content in machine-readable form. Non-compliance costs up to €15 million or 3% of global annual revenue, whichever is higher. Anthropic chose to apply this globally, not only in Europe.

Technically, Claude now uses two mechanisms: an invisible signature embedded in generated text that travels with it even when copied and pasted, and the open C2PA standard for files. This applies to the consumer app, the developer API, Claude Code, Claude Cowork, and versions available on AWS, Google Cloud, and Microsoft Foundry.

For anyone using Claude in content creation work, this matters even if you don't feel it yet. Platforms that implement AI content detectors will have the technical capability to identify text generated with Claude. That's not a problem in itself, but it does shift the conversation about transparency and authorship that many businesses and creators haven't had internally yet.

In what contexts does content generated with AI by your team or for your clients need to be explicitly disclosed?