Edit

Provenance detection overview

The Content Provenance Detection API helps determine whether digital media contains verifiable provenance information about its origin, creation, or modification. This information supports transparency, content verification workflows, and trust in non-generated, AI-generated, and digitally modified media.

What is provenance?

Provenance is information that helps identify where content came from and whether it was modified.

For example, provenance information can indicate:

  • Which application generated a piece of content.
  • Which AI model was used.
  • When the content was created.
  • Whether the content was edited after creation.
  • Whether the content contains supported authenticity signals.

Provenance information doesn't determine whether content is true, accurate, harmful, or trustworthy. Instead, it provides origin signals that applications can use as part of their own content evaluation and transparency experiences.

How the Content Provenance Detection API works

The Content Provenance Detection API can check whether media content was generated by Microsoft GenAI applications.

The API can identify:

C2PA manifests created by Microsoft

The Coalition for Content Provenance and Authenticity (C2PA) standard defines a cryptographically signed metadata format that you can embed within media files.

A C2PA manifest can contain information about:

  • Content creator.
  • Generation tools.
  • Edit history.
  • Creation timestamps.

Applications can use this information to understand the origin and lifecycle of content.

Invisible watermark signals from Microsoft

AI generation systems can embed invisible watermarking signals into generated content. These signals aren't detectable visually or audibly.

When a supported watermark signal is detected, the Detect API service can provide information about the associated content generation system.

Content Provenance Detection API Limitations

The API should not be used as a standalone determination of authenticity or trustworthiness. The following are limitations:

  • Does not determine whether content is true or false
  • Does not identify all AI-generated content
  • Does not guarantee attribution for all media
  • Does not assess content safety or harmfulness

A result indicating that no provenance information was detected does not necessarily mean that content was not AI-generated. It only means that the service did not identify supported C2PA provenance and watermarking signals.

Supported media types

The API supports media content under 100 MB. These types include image files (JPEG, PNG, GIF, WebP), audio files (MP3, WAV, MP4-audio), and video files (MP4).

Frequently asked questions

What does the Content Provenance Detection website and API do?

The Content Provenance Detection tool checks whether media content contains provenance or invisible watermarking signals that indicate it was generated by Microsoft AI systems. The API provides the tool's capabilities.

What content can I detect?

You can detect and verify whether image, audio, or video content was generated by Microsoft AI systems only.

How are provenance signals detected?

The signals are detected with C2PA Content Credentials and invisible watermarking technologies.

C2PA Content Credentials is an open technical standard where metadata about a piece of content, called the manifest, is cryptographically attached to the content itself. It provides information about how, when, and by whom the content was created or modified. The API detects and verifies this information by checking the manifest's integrity.

Invisible watermarking embeds a machine-readable signal directly into the content itself, such as the pixels of an image or the waveform of an audio file. These signals remain imperceptible to human users. The API analyzes the content for this embedded signal and can recover identifying information if the content was generated by Microsoft AI systems.

What file types are supported?

Supported file types include:

  • Image: JPEG, PNG, GIF, WebP
  • Audio: MP3, WAV, MP4-audio
  • Video: MP4

What file sizes are supported?

Files with sizes up to 100 MB are supported.

My file is larger than the 100 MB limit. Can I still detect provenance signals?

Yes. If your file exceeds the 100 MB limit, use a cropped portion of the content that's under 100 MB. For content generated by Microsoft AI systems, provenance can still be detected and verified through embedded invisible watermarking, even when only a portion of the original content is analyzed.

For audio or video files, use Clipchamp to create a smaller clip before uploading.

Related content

  • Last updated on