Anthropic launched a web tool at claude.com/check-files that lets anyone upload a file and learn whether Claude produced it. What the checker reads is a content credential, the signed provenance record Claude stamps onto supported downloads such as PNG, JPG, SVG, MP4 and WAV files.
The credential follows C2PA, an open standard also used by camera makers and photo-editing software to record an image’s origin. The record follows the C2PA standard, so other compatible tools can read it too; this one, Anthropic says, calls out credentials that trace to Claude.
The tool checks files only. Detecting Claude’s separate text watermark, embedded in drafts like essays or emails, takes a separate Detection API, which Anthropic describes as privately previewed to eligible organisations, driven by an EU legal obligation. Anthropic notes that OpenAI and Google DeepMind offer comparable checkers for their own models.
Uploaded files stay on the user’s device. Per Anthropic, the tool reads only the embedded credential and never stores the file or uses it for any other purpose.
A watermark check answers whether Claude touched a file, not whether AI made it. Content with no credential could still be AI-generated by another provider, and the whole approach works only because Anthropic chose to attach one.
Per Anthropic’s claude.com/check-files product page, published September 2026.
