Anthropic teases AI detection tools then admits watermarks can’t prove authorship or expose every AI fake

Anthropic will embed robust, machine-readable watermarks into text generated by all new Claude AI models worldwide, fully delivering on the commitments the company made under the European Union’s AI Act transparency code.

EU rules, global shockwaves. The new requirements hit every Claude model launched in the EU on or after Aug. 2, and Anthropic says it will push this watermarking system everywhere Claude operates, not just inside Europe.

• Watermarking will cover output from supported models across Claude, the Claude API, Claude Code, Claude Cowork and Claude Tag.
• Anthropic is also working to add watermarking to models released before Aug. 2 during the AI Act’s transition period.

Invisible marks in text. When a supported Claude model generates text, it embeds an imperceptible, machine-readable watermark into the response. Anthropic said the watermark does not change the meaning, quality or readability of the text.

• Because the watermark is embedded in the text itself, it stays with the content when it is copied and pasted. It may also survive some editing.
• Watermarking happens at the model level, so it is included regardless of which Claude product generated the output.
• Supported PNG, JPG and SVG files will also receive signed C2PA provenance metadata.

Yes, but. Anthropic said it will provide tools for users and third parties to detect embedded text watermarks and signed provenance metadata. Technical documentation will be released later. The company also cautioned that a detected watermark only shows that the content may have been processed by Claude.

• It doesn’t prove Claude created the original ideas or wrote the original text.
• Likewise, the absence of a detectable watermark doesn’t mean content wasn’t AI-generated.
• Older models, heavily edited text, and files with stripped metadata may not contain detectable marks.

Why you should care. Claude is widely used to create, edit, and refine marketing content. Anthropic said the watermarks are intended to improve transparency. If AI text watermarking becomes more common, it could change how AI-generated content is identified and disclosed.

Claude’s support page. How Claude marks AI-generated content