Anthropic embeds invisible watermarks in Claude models worldwide

Starting August 2, 2026, every new Claude model Anthropic ships will carry a hidden signature. The company plans to embed machine-readable watermarks into all text outputs generated by its AI, a move designed to make synthetic content identifiable even when it looks indistinguishable from human writing.

The watermarks won’t be visible to the naked eye. They’ll be baked into the statistical patterns of generated text, essentially a digital fingerprint that detection tools can read but humans cannot. Anthropic is also adding digitally signed provenance metadata for certain supported file types, creating a paper trail for AI-generated content that extends beyond plain text.

Why now, and why everyone gets it

The timing is not coincidental. Anthropic signed onto the EU AI Act’s Article 50(2) Code of Practice, a voluntary framework that becomes legally enforceable on August 2, 2026. Article 50 requires AI providers to transparently mark their generated content so that downstream users, platforms, and regulators can distinguish it from human-created material.

What’s notable is the scope. These watermarks won’t be limited to European users or EU-facing products. They’ll apply across all Claude platforms globally, including the API, Claude’s chat interface, Claude Code, and other associated tools.

Models launched before the August 2026 deadline won’t be exempt, either. Anthropic has committed to providing what it calls “transitional marking support” for older models.

The technical challenge of text watermarking

The general approach involves subtly biasing the model’s word choices during generation. Instead of picking the single most probable next word, the model might slightly favor certain tokens in a pattern that a detection algorithm can later identify. Done well, the text reads naturally. Done poorly, it introduces awkward phrasing or reduces output quality.

Anthropic hasn’t published the technical details of its specific implementation. But the company will need to convince regulators that its approach is robust enough to serve the EU AI Act’s transparency goals.

What this means for the AI industry

Anthropic is not the first company to experiment with AI watermarking. Google DeepMind developed SynthID for marking AI-generated text, images, and audio. OpenAI has discussed watermarking approaches but has been slower to deploy them broadly.

The EU AI Act applies to all providers serving European users, not just the ones that signed the voluntary code of practice early.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.