I would like to propose an idea for strengthening AI-generated image provenance and accountability.
I understand that technologies such as C2PA Content Credentials and invisible watermarks such as SynthID already provide important provenance capabilities.
My suggestion is to explore a further layer: a distributed, cryptographically protected hidden fingerprint embedded throughout the image.
Instead of placing a watermark at one fixed location, the fingerprint could be distributed across the entire image so that it has a better chance of remaining detectable after ordinary cropping, resizing, compression, screenshots, or minor editing.
The fingerprint could securely reference a generation record containing:
- AI generator
- Unique generation ID
- Date/time of creation
- Originating account/login identifier
The account information could remain private and only be accessible to the provider through an appropriate authorized process.
A verification system could then provide something like:
AI provenance verified ✓
Generator: XYZ
Creation time: [date/time]
Generation record: Verified
And importantly:
No provenance found ≠ Genuine
Instead, it should say:
“No reliable provenance found — authenticity cannot be determined.”
The reason I think this matters is that highly realistic AI images can sometimes look completely genuine. Visual examination by humans—or even another AI—can therefore produce an incorrect conclusion.
My long-term vision is:
AI generation → distributed fingerprint → secure generation record → independent verification
I would be very interested to hear what others think about the technical feasibility and whether similar approaches are already being explored.