Anthropic confirmed on 11 August 2026 that it will embed invisible watermarks in text generated by Claude, Claude Code, Claude Cowork and the Claude platform API — everywhere Claude is offered, not just in Europe. The trigger is the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content, which Anthropic has now signed; non-compliance risks fines of up to €15 million or 3% of global turnover. Every Claude model released since 2 August 2026 supports the marking from day one. For image and document files, Anthropic is adding signed provenance metadata using the C2PA standard, the same approach already used for AI-generated images elsewhere.
The mark itself is designed to be imperceptible — Anthropic says it doesn’t change a response’s meaning, quality or readability, and the model generating the content isn’t even aware it’s happening. That’s the detail worth sitting with: this isn’t a visible “AI-generated” label you can strip out by editing the text. It’s a persistent signal that survives copy-paste, and Anthropic has said a detection API is coming that anyone will be able to query. Google, OpenAI and other major labs are moving toward comparable schemes, so this is becoming an industry norm rather than an Anthropic quirk.
So what
If your team ships code, copy or content generated with Claude Code or any Claude product, there’s now a standing, checkable record that it was AI-assisted — which matters for procurement processes, client disclosure requirements, and any regulated sector where provenance of work product is scrutinised. It doesn’t change how you should be building today, but it’s another reason “AI wrote this and nobody checked it” is a worse position to be in a year from now than it is today. We build AI-assisted development processes with proper review and provenance baked in from the start — see how we approach AI-assisted development or get in touch to talk it through.