Media Provenance ExplainedEpisode 02The distinction

Media provenance vs. deepfake detection

2 min read
DETECTIONPROVENANCEchecked after the factsigned at the source

When a suspicious video appears online, many people ask the same question: is it fake? This is an important question, but it is not the only one public institutions should consider. Another question is whether they can establish a verifiable connection between the video and the institution that originally published it.

Deepfake detection and media provenance approach this problem from different directions. Detection is reactive: it examines content for indications that it may have been generated or manipulated. The output may identify suspicious features or estimate the likelihood of alteration. Such assessments can be useful, but they are not the same as a documented record of origin.

Two directions, one problem

Deepfake detection

  • Reactive
  • Examines content for signs of possible manipulation
  • Can estimate the likelihood of alteration

Media provenance

  • Proactive
  • Starts at a trusted point in the content’s lifecycle
  • Documents origin, processing and changes, signed to be verifiable

Media provenance is proactive. It begins at a trusted point in the content’s lifecycle and can document information about its origin, processing and subsequent changes. Digital signatures can make that information verifiable. The objective is not to guess whether every video on the internet is authentic. It is to provide evidence for the content an institution creates and publishes.

Imagine that an official hearing recording is posted to the institution’s archive. Later, a short excerpt circulates on social media. A detection tool may examine the excerpt for manipulation. A provenance mechanism, where supporting credentials are available, can instead help establish what source signed the material and which recorded actions form part of its history.

These approaches can complement each other. A video without Content Credentials is not automatically false, and a video with valid credentials is not automatically factually correct. Detection systems can also make mistakes. Institutions should avoid turning either approach into a simplistic ‘true or false’ label.

For an organisation responsible for publishing the public record, provenance offers a practical starting point: build verifiability into what you control. Detection may still be useful for assessing suspicious material circulating outside those controlled workflows.

The takeaway

Detection asks what may have been altered. Provenance asks what can be verified about a content item’s declared origin and history.

Tip: use ← and → to move through the series.