Anthropic introduces machine-readable watermarks for claude content

US-based artificial intelligence company Anthropic began adding machine-readable “watermarks” to text and files generated by its new Claude AI models on August 2, AzerNEWS reports.

According to Interesting Engineering, the labeling is linked to new regulatory requirements in the European Union and applies not only to users in the EU but worldwide.

The markings are not visible to users when reading the generated content. However, they may remain when Claude responses are copied and pasted and, in some cases, can survive subsequent editing.

The model-level labeling applies across Claude products, including the API, Claude Code and Claude Cowork. It also covers Claude models made available through cloud platforms such as AWS, Google Cloud and Microsoft Foundry.

The measure applies only to new Claude models. Previously released versions will not receive the new markings.

For files generated by Claude, Anthropic uses the C2PA standard, which is designed to provide information about whether Claude was involved in processing the material and to track subsequent changes to its provenance data.

Anthropic stresses that the presence of a watermark does not necessarily mean that the content was authored by AI. Claude can also be used to process or modify material that was originally created by a human.

At the same time, the absence of a watermark does not prove that content was created by a person. Extensive editing or a small amount of text can weaken the marking or make it undetectable.

The move represents another step toward making AI-generated and AI-assisted content more traceable, particularly as regulators and technology companies seek ways to improve transparency around the origin and processing history of digital content.