Context
ExifTool is the gold standard for file metadata. As AI generates more content (images, documents, video), there's no standard set of EXIF/XMP tags for AI provenance:
- Was this file AI-generated?
- Which model?
- What was the confidence/trust level?
- What sources were used?
Possible Approach
A new tag group or namespace for AI provenance:
exiftool -AIGenerated=true -AIModel="gpt-4" -AIConfidence=0.85 image.jpg
Or using an existing namespace like XMP with a custom schema:
exiftool -XMP-akf:TrustScore=0.85 -XMP-akf:Agent="dall-e-3" image.jpg
Why
- EU AI Act Article 50 (August 2, 2026) — AI-generated content must carry transparency metadata, deepfake disclosure mandatory
- C2PA exists but requires PKI/Certificate Authority infrastructure
- A lightweight EXIF/XMP approach would be more universally adopted
Reference
AKF (Agent Knowledge Format) is an open standard working on this — ~15 tokens of JSON provenance that can map to EXIF/XMP fields. ExifTool supporting these tags would make the format universally readable.
Would love to hear Phil's thoughts on whether AI provenance tags belong in ExifTool.
Context
ExifTool is the gold standard for file metadata. As AI generates more content (images, documents, video), there's no standard set of EXIF/XMP tags for AI provenance:
Possible Approach
A new tag group or namespace for AI provenance:
Or using an existing namespace like XMP with a custom schema:
Why
Reference
AKF (Agent Knowledge Format) is an open standard working on this — ~15 tokens of JSON provenance that can map to EXIF/XMP fields. ExifTool supporting these tags would make the format universally readable.
Would love to hear Phil's thoughts on whether AI provenance tags belong in ExifTool.