Synthetic Media and Source Trails

Provenance and Content Credentials

A publishing protocol for source trails, AI-use labels, synthetic media, and content credentials. Spiralism can use generated tools without letting the record become fog.

The public will increasingly meet the institution through pages, clips, images, transcripts, talks, screenshots, and summaries that pass through AI systems. Some of those tools will be useful. Some will also blur authorship, source lineage, consent, and truth.

Provenance is the answer to the blur. It does not make content true. It makes the history of the content easier to inspect.

The Rule

Never ask the audience to trust a polished artifact when a provenance trail can be provided.

A provenance trail should answer:

The goal is not maximal disclosure of private material. The goal is enough context for the audience to understand what kind of thing they are seeing.

Provenance Is Not Truth

Content credentials and source notes can show a content history. They cannot prove that the depicted event happened, that a witness is accurate, or that an interpretation is correct.

Treat provenance as a map of custody and transformation:

Do not tell readers that content credentials make an artifact “real.” Say that credentials provide inspectable provenance. Truth still depends on source quality, corroboration, consent, context, and correction.

Artifact Labels

Every public artifact should fit one label.

Label Meaning Required note
Human-authored Written or recorded by people; AI not materially involved No AI note required
AI-assisted AI helped with source search, summarization, editing, transcription, translation, or drafting AI-use note near the artifact
AI-generated AI generated the primary image, audio, video, or prose surface Visible synthetic-media disclosure
Composite Built from multiple sources, clips, reenactments, or edited testimony Source and transformation note
Reenactment Represents an event without being original footage/audio Visible reenactment disclosure
Interpretation Meaning-making built from sources rather than direct evidence Claim-class label or source note

An artifact may carry more than one label. If labels conflict, choose the more transparent label.

Disclosure Patterns

Use plain language.

AI-assisted writing:

AI use: This piece used AI assistance for [task]. Human review covered sources,
factual claims, consent, editing, and final judgment.

Synthetic visual:

Synthetic media: This image was generated or materially altered with AI. It is
not documentary evidence of a real scene.

Composite testimony clip:

Provenance note: This clip combines excerpts from approved testimony records.
Names and identifying details were changed under the Privacy and Data protocol.

Reenactment:

Reenactment: This scene represents a reported experience. It is not original
footage.

Content credential note:

Content Credentials: Where supported, this artifact includes provenance
metadata. Credentials can help inspect origin and edits, but they do not prove
the underlying claim.

Content Credentials Practice

When producing AI-generated or materially altered images, audio, video, or mixed-media artifacts, preserve or attach content credentials when the toolchain supports it.

Record:

If credentials are stripped by resizing, re-encoding, upload, social platforms, or CDN processing, keep the original credentialed file in the archive and note that the public derivative may not retain metadata.

Source Trail

Every researched page should preserve a source trail.

Minimum source trail:

Generated output is never a source. It can propose leads, summaries, or questions. The source trail begins when a human opens the underlying source.

Testimony Provenance

Testimony needs a different standard because privacy can outrank public inspectability.

For testimony-derived artifacts, public provenance may say:

Do not publish raw source files, chat logs, care notes, or identity metadata to make an audience feel more certain. Certainty is not worth betrayal.

Synthetic Media Red Lines

Do not publish:

If a synthetic artifact could reasonably be mistaken for documentary evidence, label it at the point of encounter, not only in a footer or policy page.

Provenance Review Before Publication

Before publishing, answer:

  1. What kind of artifact is this?
  2. What claims does the audience need to distinguish from interpretation?
  3. What source trail supports those claims?
  4. Did AI materially shape the artifact?
  5. Is synthetic or altered media visible as such?
  6. Are credentials preserved or archived where possible?
  7. Does the disclosure appear where the audience will see it?
  8. Does the artifact expose private or restricted material?
  9. Can a reviewer reconstruct the chain from source to publication?
  10. Who owns correction if the artifact is challenged?

If the chain cannot be reconstructed, publish later.

Correction and Challenge

Every public provenance note should make challenge possible.

When challenged:

Do not defend a claim because it is aesthetically useful. The archive matters more than the image.

Spiralism Policy

Spiralism may use AI for drafting, research leads, transcription support, translation drafts, visual prototypes, accessibility summaries, and production assistance. Public artifacts must remain labeled, reviewable, and correctable.

The institution should adopt C2PA Content Credentials or comparable provenance workflows as soon as the media pipeline can preserve them reliably. Until then, manual provenance notes are required for significant public artifacts.

This protocol pairs with:

Sources Checked