Anthropic’s EU Code Commitment Brings Watermarking Questions Into Focus

Anthropic's decision to sign the EU's General-Purpose AI Code of Practice adds to industry trends towards transparency around AI-generated content. However, Anthropic has not yet released broad, live watermarking features for its Claude text outputs or published detailed customer implementation terms. The company's signing of the EU framework represents its engagement with governance, safety, and transparency guidelines for general-purpose AI.

The immediate concern for Claude users is the unclear division of compliance responsibilities between the model provider and organizations deploying AI-generated content. While Anthropic's participation in the regulatory process is confirmed, the operational scope, format, and availability of Claude watermarking remain open questions.

The EU code aims to help AI model providers demonstrate compliance with EU AI Act obligations, particularly regarding AI-generated content identification through labeling, watermarking, and provenance approaches. The code provides an implementation framework rather than a single technical standard, leaving room for different methods across providers, modalities, and customer workflows.

Anthropic's Transparency Hub has noted that watermarking is an area of ongoing discussion and that Claude does not currently generate image outputs or provide watermarking for its text outputs. Enterprise AI teams should view watermarking as just one part of a broader governance program, which includes human oversight and clear accountability.

Businesses deploying generative AI should track official provider documentation, map high-risk content workflows, and establish disclosure processes independent of a single detection method. While provenance technology can be useful, it is not a substitute for comprehensive governance.