SiftVeil

Content Credentials

Proof when proof exists.

SiftVeil can inspect supported media for Content Credentials — signed data designed to travel with a media item and record where it came from and how it was edited. Where that evidence is there, SiftVeil reads it locally. Where it is not, SiftVeil says so and stops.

BetaIncluded in SiftVeil 0.4 on both the Free and Pro plans, where supported.

Missing Content Credentials do not mean content is human-made.

Social platforms may remove provenance metadata when media is uploaded or transformed, and most media never carried credentials to begin with. SiftVeil reports what it can verify and stays quiet about what it cannot.

What Content Credentials are

Content Credentials are a way of attaching tamper-evident information to a media file that describes where it came from and what has happened to it. They are built on C2PA, an open industry standard developed by a coalition of camera makers, software vendors, publishers and platforms.

When a tool supports the standard, it can sign a record into the file — for instance that an image came from a particular camera, or that a generative AI tool produced or modified it. Because that record is cryptographically signed, later alteration can be detected.

Two things follow from how the standard is designed:

  • Credentials are only useful when they are actually present. Nothing compels a tool to write them, or a platform to preserve them.

  • Credentials describe handling, not truth. A file with flawless provenance can still be presented misleadingly.

What SiftVeil checks

  • Looks for Content Credentials attached to supported media
  • Checks whether the credential data validates
  • Looks for assertions relating to generative AI, such as AI-generated or AI-edited
  • Records the outcome as a local numeric counter
  • Applies whatever rule you chose for that outcome

What it does not do

  • Analyse pixels, audio or writing style to guess whether something is AI
  • Send your media to a server for analysis
  • Assign a confidence score to content that carries no credentials
  • Treat a missing credential as a signal of any kind

Result states

Every outcome, and what it isn’t

Each state is described twice — what SiftVeil observed, and what that observation does not entitle anyone to conclude.

  • Verified AI-generated

    The item carries valid Content Credentials asserting that it was created with a generative AI tool.

    What it does not mean: That every AI-generated item in your feed will be labelled this way. Most media carries no credentials at all.

  • Verified AI-edited

    Valid credentials record that a generative or AI-assisted editing step appears in the item’s history.

    What it does not mean: That the item is wholly synthetic, or that the edit was deceptive. Ordinary editing tools can produce this assertion.

  • Content Credentials found

    The item carries valid provenance metadata describing its origin or edit history, with no AI assertion that SiftVeil recognises.

    What it does not mean: That the content is accurate or trustworthy. Provenance describes how a file was handled, not whether it tells the truth.

  • Invalid or unverified credentials

    Provenance data was present, but SiftVeil could not validate it — it may be incomplete, altered, or signed by an issuer SiftVeil cannot check.

    What it does not mean: That someone tampered with the item. Routine re-encoding by a platform breaks signatures all the time.

  • No provenance found

    SiftVeil found no Content Credentials attached to this item.

    What it does not mean: That the item is human-made, authentic, or not AI. This state carries no information about how the media was created.

  • Unsupported or skipped

    The item’s format or delivery is outside what Provenance Beta inspects, or the check was skipped so browsing stays responsive.

    What it does not mean: Anything at all about the content. A skipped check is simply an absence of information.

How a check runs

One item, start to finish

The same path every time, with an exit at any point where the evidence runs out.

  1. A media item appears in your feed
  2. Local Content Credentials check
    Runs on your device
    • Credentials foundValid, readable provenance
    • No credentialsTells you nothing either way
    • Unsupported or skippedCheck not attempted
  4. SiftVeil applies your selected rules where reliable evidence existsIf a check fails or times out, SiftVeil fails open and leaves the item alone.

Limitations

Why SiftVeil can’t verify most of your feed

These are not temporary gaps we expect to close. They are properties of how media travels across the internet.

Most media carries no credentials

Signing is opt-in and relatively new. The overwhelming majority of images and video online has never carried Content Credentials, whoever or whatever made it.

Platforms strip metadata

Social platforms routinely discard metadata when they process an upload. Provenance that existed on the original file is often gone by the time it reaches your feed.

Re-encoding breaks signatures

Compression, resizing, cropping and format conversion can invalidate an otherwise genuine credential. An invalid result usually means ordinary processing, not tampering.

Coverage is partial

Not every media format, player or delivery method is supported. Where SiftVeil cannot inspect an item, it reports the check as unsupported rather than inventing a result.

Checks are skipped when they would get in the way

Provenance checking is designed not to interfere with normal browsing. If a check would be slow or cannot complete, SiftVeil fails open and leaves the item alone.

Provenance describes handling, not truth

A perfectly signed photograph can still be captioned misleadingly. Credentials tell you about a file’s history — not whether what it shows is accurate.

SiftVeil does not infer that media is human-created merely because Content Credentials are missing.

This is the single most important thing to understand about Provenance Beta. A missing credential is not a clean bill of health, a signal of authenticity, or evidence of anything at all. It means the file arrived without a record — which is the normal case for almost everything you will scroll past today. Anything that claimed otherwise would be selling you certainty that does not exist.

FAQ

Provenance questions

The questions worth asking about a feature like this.

So does this tell me whether something is AI?
Sometimes, and only in one direction. If a media item carries valid credentials asserting that generative AI was involved, SiftVeil can tell you that. If it carries nothing, SiftVeil cannot tell you anything — and it will not pretend otherwise.
What does it mean when a video has no credentials?
Only that no credentials were found. It is not evidence that the video is real, human-made, authentic or unedited, and it is not evidence of the opposite either. It is an absence of information.
Does provenance checking slow down my browsing?
It is designed not to. Checks are best effort and fail open: if one would be slow, cannot complete, or hits unsupported media, SiftVeil abandons it and leaves the item untouched rather than holding up your feed.
Is any of my media sent somewhere to be checked?
No. Where provenance inspection is supported, it happens locally on your device. SiftVeil does not upload your feed media for provenance analysis.
What does SiftVeil keep about these checks?
Local numeric counters — how many items were checked and how many landed in each result state. They are aggregates held in your browser, not a record of which items you saw.
Why is it labelled Beta?
Because coverage is incomplete and the ecosystem is still young. Formats, platform behaviour and the standard itself are all changing. Calling it Beta is a statement about how much you should rely on it, not a marketing flourish.

Provenance Beta is one signal among the filters you configure. For how the rest of SiftVeil handles your data, see the provenance section of our Privacy Policy, and for the limits we commit to in writing, the provenance limitations in our Terms.

Filter on evidence, not vibes.

SiftVeil applies your rules where there is something real to act on, and leaves the rest of your feed alone.

Get SiftVeilCompare Free and Pro

Privacy-first · Local filtering · Free plan available · Currently in paid beta