Artificial intelligence company Anthropic has announced plans to introduce watermarking for text generated by its AI models, including Claude, as it moves to comply with new European transparency requirements.
The company said models released after August 2 will automatically incorporate technology designed to identify AI-generated text and files.
The change follows the introduction of new transparency requirements under the European Union’s AI regulatory framework, which seeks to make AI-generated or manipulated content more readily identifiable.
For generated files, Anthropic said it will use the C2PA open standard, which enables information about the origin and editing history of digital content to be embedded in files.
Anthropic said the text watermarking technology will operate at the model level, meaning it will apply regardless of which Claude product or interface generates the content. The company also plans to extend the technology to older models.
According to Anthropic, the watermark is designed to remain associated with generated text when it is copied and pasted and could survive some subsequent editing.
“Because the watermark is part of the text, it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing,” the company said.
Questions remain, however, about the durability of text watermarks, including how extensively AI-generated material would need to be edited before the identifying signal could no longer be detected.
Anthropic has yet to provide detailed information on how resistant its system will be to extensive rewriting or other modifications.