Anthropic's Claude to Include “Invisible Watermarks” In Its AI-Generated Content That Can Bypass Copy-Pasting and Editing

IF YOU’VE BEEN USING Claude, the Artificial Intelligence (AI) chatbot, for skirting the rules for your class assignments and copy-pasting text from the chatbot to bulk up your essays, then we’ve got bad news for you. Recently, Claude’s parent company Anthropic announced that any content that the AI chatbot generates from now onwards will be marked by invisible watermarks and metadata.

Anthropic has recently updated its support page for marking AI-generated content. The tech company clarified that it is adding machine-readable markers to content generated by new Claude models launched after August 2, 2026, as it moves to comply with transparency requirements under Article 50 of the European Union's AI Act. 

This means, from now onwards, everything created by Claude — be it text, images, or processed files — will carry a watermark that flags it as AI-generated. One can’t see these markers, nor would it change how the text reads. But Anthropic said that Claude can embed an imperceptible, machine-readable watermark directly into AI-generated texts and other content. As this watermark is part of the text itself, it can travel with the text when it is copied and pasted elsewhere, and may also survive some editing.

See also: Anthropic Says Its Claude AI Broke Into Three Organisations During a Private Security Test

The watermarking will apply across all of Claude's products, including the chatbot app, the API, Claude Code, Cowork, and Tag, as well as versions of Claude used through Amazon Web Services (AWS), Google Cloud, and Microsoft. 

This decision by Anthropic comes as concerns emerge regarding the rise of “AI slop content” — mass-produced, low-quality AI-generated content — which has saturated the current creative market. Mainstream concerns point towards being unable to segregate AI-generated and non AI-generated content. Anthropic’s latest announcement has provided a concrete way to identify whether any content is organically produced, or improved, or generated via AI. 

How do Claude’s invisible watermarks work?

Anthropic says that these invisible watermarks will work in two ways. For AI-generated text, a supported Claude model will embed the watermarks directly onto the text itself. The tech company said these watermarks cannot be detected and removed by users, but it will remain in place even if one copy and paste the text to a document, and may also “persist through some editing.” 

See also: Chinese AI firms threaten national security through illegal extraction from US companies: Anthropic

However, these watermarks will not impact the quality or meaning of Claude’s responses, Anthropic continued. 

For different file types, such as .svg, .png and .jpg, Claude will attach “signed provenance metadata” to these files. This metadata follows the Coalition for Content Provenance and Authenticity (C2PA) open standard, which is used across the tech industry to record information about content provenance. If a signed metadata label is present, it signals that a file was processed by Claude and lets you detect whether the file has been tampered with.

That being said, the two types of watermarks aren't equally effective. Text watermarks are built to survive copy-pasting and even some editing. File watermarks, though, are more fragile — Anthropic says they can be wiped out just by converting the file format, saving it again, or even taking a screenshot. So if a file doesn't show a watermark, that doesn't necessarily mean it wasn't made by AI — it might just mean the watermark didn't survive. 

Anthropic says that it is working to enable users and other third parties to detect Claude's embedded watermarks and provenance metadata. The tech company stated further details on detection mechanisms will be revealed in future technical documentation.

(Edited by Agniva Ray)

Suggested reading: