Industry Odisha Bureau, Aug 11: In latest update, Anthropic has started adding watermarks to content generated by AI platform Claude. What makes it interesting is that the watermarks will be invisible, machine readable only. This marks a major step towards making AI generated texts and files easier to identify.
The company recently updated its support documentation on AI-generated content. It said that the system is being introduced as part of its efforts to comply with transparency requirements under Article 50 of the European Union’s AI Act.
For text, supported Clause models will embed an invisible watermark directly into their output. Users will not be able to see the mark. However, Anthropic says it can remain intact when text is copied into another document and may survive some editing. The watermark does not alter the meaning or quality of the generated response.
For files including PNG, JPG and SVG, Claude will attach signed provenance metadata based on the Coalition for Content Provenance and Authenticity (C2PA) standard.
Anthropic further says that it is developing detection mechanisms that will allow users and third parties to identify these marks. However, the company cautions that detecting a watermark is not conclusive proof that Clause originally created the content. Users may employ Claude for proofreading, translation, summarisation or other forms of editing.
Similarly, content can still be AI generated even if it lacks watermark. Content produced by older models might not have the marks. The same is applicable to materials that havebeen heavily edited, paraphrased or combined with other text. The marking system will apply to supported Claude models worldwide across Claude, Claude Code, Claude Cowork, and other platforms. Anthropic’s approach to watermark Claude content is similar to Google’s SynthID, which also uses invisible markers to help identify AI-generated content.
