Anthropic starts watermarking AI content in the EU

In August 2026, Anthropic announced that its new AI models would begin watermarking content in the European Union as part of compliance with the EU AI Act. This legislation, which imposes strict obligations on AI systems, aims to enhance transparency by requiring that users are informed when they are interacting with AI-generated content. The new models, launched on or after August 2, will support machine-readable content marking, which includes digitally signed provenance metadata and embedded watermarks. These watermarks will be accessible only through specific cloud partners such as Amazon Web Services, Google Cloud, and Microsoft Foundry.

The Claude-marker, a part of Anthropic's initiative, is one of over 200 signatories of the EU AI Act's Code of Practice on Transparency of AI-Generated Content. This voluntary tool is designed to assist businesses in adhering to the new regulations. Other major AI providers, including OpenAI, Mistral, Meta, and Microsoft, have also signed this code, indicating a collective effort to promote transparency in AI-generated content.

Despite the advancements in watermarking technology, Anthropic acknowledges several challenges. For instance, content may not carry a watermark if it has been heavily edited, paraphrased, or translated. Additionally, if the generated text is too small or if the file's metadata is stripped through conversion or screenshots, the watermark may not be present. There is also a possibility that the watermark could appear on human-generated content if it has been processed by Claude for tasks like reading or summarizing.

As AI technology continues to evolve, the need for effective labeling mechanisms becomes increasingly critical. The rise of AI-generated content has led to concerns about the authenticity and integrity of information. In response, countries like China have already implemented laws requiring social media companies to label all AI-generated content. Anthropic's efforts to watermark its AI-generated content represent a significant step towards ensuring that users can identify and verify the origins of the information they consume, thereby fostering trust in AI technologies.