Claude Content Credentials

Claude’s generated files and Claude’s text watermark are separate systems. The text watermark is a keyed pattern in token choice. Content Credentials are signed C2PA metadata attached to supported files.

What Anthropic documents

Anthropic says that when Claude produces a supported file type such as PNG, JPG, or SVG, it attaches a cryptographically signed note in the file’s metadata stating that the file was made or processed with Claude. The Help Center names the Coalition for Content Provenance and Authenticity (C2PA) standard. Anthropic also notes that the credential is not the same as the text watermark: nothing in the pixel payload has to change for the label to be present.

What credentials communicate

If a signed label is present and readable, it can signal that Claude processed the file and, depending on the validator, whether that binding still verifies. Anthropic says the label does not include identifying information about a user.

What they do not communicate

• They do not prove the depicted scene is true.
• They do not prove an image is or is not “AI art” in a moral sense.
• They do not replace the statistical text watermark.
• Absence of credentials does not prove a file was not processed by Claude — metadata can be stripped.

How to inspect them here

The Image Metadata & C2PA Inspector reads embedded Content Credentials in the browser for JPEG, PNG, and WebP. It does not currently inspect SVG, HEIC, or TIFF. Anthropic’s documented SVG output is therefore outside this inspector’s support matrix even if Claude attached credentials.

Trust-list verification is not enabled in the current inspector. Finding credentials is not an authenticity score.