Anthropic to watermark Claude AI content globally under EU AI Act

Dublin Ireland july 20,2026..a computer screen with claude website displayed.. — Image credit: Search Engine Journal

Anthropic announced Tuesday it will embed invisible watermarks and signed provenance metadata into content generated by its Claude AI models worldwide, aligning with European Union transparency regulations.

The artificial intelligence developer signed the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content, committing to global implementation of content marking for its generative AI outputs.

These markings will indicate potential processing by Claude models, rather than claiming authorship, and will apply to content produced across various Claude applications and cloud platforms, the company said.

Text watermarks are integrated at the model level and are designed to persist through copying and some forms of editing. File metadata will adhere to the C2PA open standard, a technical specification for content provenance.

Anthropic stated that content may be marked even if original ideas are human-generated, particularly when Claude is used for tasks such as proofreading, translation, summarizing or file conversion.

However, the effectiveness of watermarks in resisting sophisticated editing remains a subject of debate among experts. Older models, extensive editing or very short text segments might not produce a detectable mark, the company acknowledged.

Roger Montti, a technology journalist at TechCrunch, noted that while watermarking is a step toward transparency, its resilience against deliberate alteration is a key challenge.

Alex Cui, co-founder of GPTZero, an AI detection platform, highlighted the ongoing technical difficulties in creating watermarks that are both effective and imperceptible. Jonas Geiping, a researcher at the ELLIS Institute Tübingen, has also published work discussing the limitations of current watermarking techniques.

Anthropic’s initiative follows similar efforts by other major AI developers, including Google, Meta, Microsoft, Mistral and OpenAI, many of whom have also committed to the EU AI Act’s transparency provisions.

The company’s Claude models are accessible through its own applications, as well as cloud platforms like Amazon Web Services (AWS), Google Cloud and Microsoft Azure’s AI Foundry.

The European Commission has emphasized the importance of such transparency measures to build public trust and enable users to distinguish between human-created and AI-generated content.

Source: Search Engine Journal