Research Center

Reproducible, browser-local tests for AI media provenance signals. Each note records the sample source, license, parser version, expected signals, actual output, misses, limitations, and steps to reproduce.

Direct definition

What is an AI media provenance test?

An AI media provenance test checks evidence carried by the original file—such as C2PA Content Credentials, PNG text chunks, and MP4 container metadata—against a documented parser and expected result. It does not decide authenticity from appearance alone, and missing metadata is not proof that a file is human-made.

Current dataset

Published evidence snapshot

Download experiment registry (JSON) →
Experiment records
5
Tested sample instances
6
File formats
JPEG, MP4, PNG
Latest test date

Counts come directly from the published experiment registry. They are not estimates of platform-wide coverage.

Quarterly synthesis

AI Media Provenance Report, 2026 Q3

The stable quarterly report compares every published record, states the dataset cutoff, separates cryptographic validity from trust, and preserves the limits of this small fixture set.

Read the Q3 report →

Comparison table

Platform, format, parser, date, and result

AICheck365 experiment registry: 5 records and 6 tested sample instances
Platform or fixture Format Samples Parser Test date Main finding
Adobe Firefly JPEG 1 aicheck 0.2.0 (c2pa-rs 0.82, wasm) Valid C2PA Content Credentials with digitalSourceType=trainedAlgorithmicMedia; the signature is cryptographically valid while the signing certificate is untrusted in the browser trust list.
OpenAI (ChatGPT / DALL·E) PNG 1 aicheck 0.2.0 (c2pa-rs 0.82, wasm) A C2PA claim (trainedAlgorithmicMedia) is present, but the signing certificate is expired, so validation returns Invalid: the provenance claim cannot be verified.
ComfyUI / Stable Diffusion XL PNG 1 aicheck 0.2.0 (c2pa-rs 0.82, wasm) No C2PA signature. PNG tEXt 'prompt' and 'workflow' chunks carry the full generation graph; the detector flags AI-related metadata at low confidence (unsigned).
C2PA conformance images JPEG 2 aicheck 0.2.0 (c2pa-rs 0.82, wasm) An image with no manifest reads as unsigned (no provenance); a deliberately tampered manifest fails with assertion.dataHash.mismatch and reads as invalid.
C2PA-signed MP4 (generic) MP4 1 aicheck 0.2.0 (c2pa-rs 0.82, wasm) A BMFF C2PA manifest validates (generic CreativeWork, not AI). The detector surfaces the C2PA signal; the AI-tool-name heuristic produced a spurious match against manifest bytes, which is a logged limitation.

Method

How these tests are run

  1. Use the named public fixture or documented sample file without uploading it to AICheck365.
  2. Run the browser-local parser version and environment recorded in the experiment note.
  3. Compare expected signals with the parser’s actual state, fields, and validation issues.
  4. Publish misses, source license, known limitations, and exact reproduction steps with the result.

Read the full detection methodology →

Limits

What this dataset cannot establish

  • The registry is a small fixture set, not a representative benchmark of all exports from any platform.
  • A valid signature does not automatically mean the signer is in the browser-local trust list.
  • Unsigned or stripped metadata does not prove human creation.
  • Parser, certificate, and trust-store updates can change validation output.
  • Editable metadata and heuristic matches require lower confidence and manual review.

Record-level evidence

Experiment notes

JPEG 2026-06-19

Adobe Firefly

Valid C2PA Content Credentials with digitalSourceType=trainedAlgorithmicMedia; the signature is cryptographically valid while the signing certificate is untrusted in the browser trust list.

Samples
1
Parser
aicheck 0.2.0 (c2pa-rs 0.82, wasm)
License
MIT via contentauth/example-assets
Read experiment →
PNG 2026-06-19

OpenAI (ChatGPT / DALL·E)

A C2PA claim (trainedAlgorithmicMedia) is present, but the signing certificate is expired, so validation returns Invalid: the provenance claim cannot be verified.

Samples
1
Parser
aicheck 0.2.0 (c2pa-rs 0.82, wasm)
License
MIT via contentauth/example-assets
Read experiment →
PNG 2026-06-19

ComfyUI / Stable Diffusion XL

No C2PA signature. PNG tEXt 'prompt' and 'workflow' chunks carry the full generation graph; the detector flags AI-related metadata at low confidence (unsigned).

Samples
1
Parser
aicheck 0.2.0 (c2pa-rs 0.82, wasm)
License
Permissive repository license via comfyanonymous/ComfyUI_examples
Read experiment →
JPEG 2026-06-19

C2PA conformance images

An image with no manifest reads as unsigned (no provenance); a deliberately tampered manifest fails with assertion.dataHash.mismatch and reads as invalid.

Samples
2
Parser
aicheck 0.2.0 (c2pa-rs 0.82, wasm)
License
Apache-2.0 / MIT via contentauth/c2pa-rs fixtures
Read experiment →
MP4 2026-06-19

C2PA-signed MP4 (generic)

A BMFF C2PA manifest validates (generic CreativeWork, not AI). The detector surfaces the C2PA signal; the AI-tool-name heuristic produced a spurious match against manifest bytes, which is a logged limitation.

Samples
1
Parser
aicheck 0.2.0 (c2pa-rs 0.82, wasm)
License
Apache-2.0 / MIT via contentauth/c2pa-rs fixtures
Read experiment →

Use the evidence

Inspect a file or learn the signal types