Does an AI image’s C2PA label survive social media?

An AI image leaving your tool with C2PA Content Credentials is not automatically the same labelled file your audience receives. Before publication it may pass through a design tool, CDN, online shop or social platform. Any of those steps can create a new file and remove its cryptographic label.

This matters for the technical side of Article 50(2) of the EU AI Act. The provision calls for machine-readable marking for AI-generated or manipulated content; a visible watermark is not a replacement. This is not legal advice. The practical question here is simpler: is the label present at origin, and can you show what happens to it in your real workflow?

What is a C2PA label?

C2PA Content Credentials are a cryptographically bound provenance record for a file. They can state which tool created it and which edits followed. They are not merely an EXIF field and not a logo placed on an image.

That distinction matters when a file is recreated. If a platform recompresses a JPEG, or a CDN makes a smaller version, the original cryptographic binding may no longer be valid. The new image can look identical while no longer carrying the same verifiable provenance.

Which platforms usually preserve it?

LinkedIn and TikTok usually preserve C2PA labels better than many publishing paths. Meta services can read the label and may show viewers an “AI info” indicator. Facebook and Instagram, however, often remove the C2PA label on upload.

Platform behaviour can change. It is therefore risky to build a marking process around one service continuing to handle files in the same way. It is also easy to confuse a platform showing its own indicator with the downloadable or re-shared file retaining the original proof.

Where labels disappear before social media

The platform is often not the first break. Common examples are:

1. A designer opens the AI image and exports a new JPEG or WebP.
2. A website or e-commerce system creates responsive image sizes.
3. A CDN changes quality, dimensions or format.
4. A user downloads the file and brings it into another system.
5. The image is placed into a PDF, slide deck or video edit.

Those are normal business steps. If they produce a new file, test whether provenance survives, is correctly updated, or whether the new output needs a new label. It is not enough that the original generator “supports C2PA”. For example, OpenAI images can contain a C2PA label at origin, but it can disappear in later processing.

A simple workflow test

1. Save one file directly from your AI tool.
2. Check whether it has Content Credentials.
3. Run the same file through the route you actually use: design, web delivery, CDN, download or platform upload.
4. Compare the final file with the original.
5. Record where the label disappears or changes.

Start with a real example rather than a policy document. That gives you a useful technical map of where marking is present, where it breaks, and what needs changing.

What if the label does not survive?

The aim is not to force every platform to preserve a file unchanged. You do not control that. The aim is to put marking into the workflow that produces your output, so source files are marked, relevant outputs are demonstrable, and points of change are mapped.

AI Detective maps where marking is absent or breaks and implements the technical part in your workflow. It is not an AI detector and does not give legal advice. A lawyer can provide a legal assessment when needed.