Anthropic Marks Content Created by Claude

Anthropic is introducing a new system for marking content created or processed by Claude. New Claude models will be released with the marking feature built in, while the company is currently working to add it to older models. The markers will apply to supported Claude content worldwide, including material created through the API and Claude apps.

Claude will use different markers for text and files. An invisible watermark will be embedded in generated text at the model level, so it will remain in the text regardless of where Claude is used. Supported files, including PNG, JPG, and SVG images, will receive signed information about their origin. The system uses the C2PA standard and helps verify whether this information has been changed.

Anthropic is also developing ways for users and third parties to detect Claude markers. The company will provide more details on how the detection system works in future technical documentation.

The markers have some limitations. A detected marker means that Claude may have processed the content, but it does not prove that the AI created it. For example, Claude may have been used only to translate or edit material created by someone else. A marker can also disappear after extensive editing, translation, file conversion, or a screenshot, while very short text may not contain enough information for reliable detection.