Microsoft's Content Provenance Detection: what you can check on a file

If you want to know whether a file carries provenance data, Microsoft offers a check. The Foundry provenance page, read on 2026-10-05, says there is a detection website and an API under the name Content Provenance Detection, tied to the C2PA Content Credentials and invisible watermarks that Foundry's supported models add.

I could read the page's description of the feature. I could not read the detection tool's own documentation, so this post sticks to what the page says and does not describe the tool's inputs, size limits or output format.

What the check can and cannot settle

Using it on a file from another generator

Microsoft's page ties the check to its own models and marks. A file from a different API may carry other credentials or none, so a negative result there is weak evidence. A positive result is stronger, but only for the signal Microsoft knows how to read.

For an image from the Sume image API, which lists models such as black-forest-labs/flux.2-pro, Sume's docs say nothing about provenance. Run the detection on the delivered file, not on the provider's example, and write down what it returned and when.

What to write in your own records

Microsoft says customers remain responsible for their own transparency obligations, so a clean check is not a disclosure. If you are labelling an ad or a social clip, use the platform's AI setting regardless. Sume's docs describe a metadata field on video, image and music jobs that Sume stores on the job and does not send to the provider. I found nothing in the docs about Sume adding, keeping or removing C2PA credentials, watermarks or platform labels, so treat that as undocumented and check the delivered file.

• The Sume job id, model id and the date you ran the check.
• The result in plain words: found, not found, or not checked.
• The exact file you checked, by name and size, because a re-export changes the result.
• Your disclosure decision, separate from the check result.