How digital provenance fails in practice, and what it takes to make authenticity verifiable again.
© 2025 Creativ Solutions BV
In October 2025, The Washington Post conducted a simple but telling experiment. Their reporters generated a short video using OpenAI’s Sora, a tool that can create realistic scenes entirely from text prompts. The video was properly labeled with C2PA Content Credentials — the industry’s chosen standard for signaling how a piece of media was produced, by whom, and with what tools.
The team then uploaded the same file to eight major social platforms. Every one of them stripped the credentials. The tags that were supposed to identify the content as AI-generated vanished in the process of uploading, re-encoding, and sharing. Only one platform added its own disclosure, and even that was buried deep in a description field, invisible to most viewers.
In other words, the provenance of the file — the one element that was meant to make the system transparent — was the first thing to disappear.
This small test captured something fundamental about the state of digital authenticity. Our mechanisms for proving origin rely on cooperation from every service that touches a file. They assume that metadata travels intact through a chain of platforms, applications, and compression algorithms that were never built to preserve it. When that assumption fails, the proof collapses.
Metadata lives next to the content. It is an attachment, not an intrinsic property.
Whenever a platform optimizes a file — reducing size, adjusting color, trimming length — the metadata often goes with it. That isn’t necessarily sabotage; it’s the natural result of efficiency-driven pipelines.
The industry’s faith in metadata mirrors an older era of computing, when a document and its label were inseparable. In today’s networked reality, the label is the most fragile part of the chain. A watermark can be cropped, a JSON block can be deleted, and a credential can be ignored.
If the proof of authenticity depends on that attachment, then authenticity itself is conditional — it exists only as long as nothing changes.
At Creativ Solutions, we approach the question differently. We design systems in which proof does not depend on attachment, preservation, or cooperation. We design systems where proof can be reconstructed from the content itself.
A deterministic approach begins from a simple physical insight: every digital artifact — regardless of how it is compressed, encoded, or transmitted — contains structural properties that can be measured and mathematically characterized.
Instead of attaching external provenance data, deterministic verification analyzes the signal itself: the internal rhythm, distribution, and timing patterns of the file. From these, it derives a set of invariants that remain stable across formats and transformations.
These invariants act as a structural signature — a mathematical fingerprint that can be recomputed anywhere, anytime, directly from the content. Even if all metadata is stripped, the underlying structure still ties back to its original proof. Verification becomes reconstruction, not preservation.
In practice, such systems do not need to trust the platform. They simply measure reality.
The Washington Post’s result was not a failure of the C2PA standard. It was a demonstration of a deeper truth: digital networks are not provenance-preserving environments. They are lossy, mutable, and optimized for delivery rather than integrity.
Expecting proof of origin to survive inside such an environment is like expecting a paper label to stay intact through a storm. The right question is not how to make the label stronger, but how to make the object identifiable without it.
That is what deterministic verification achieves. It shifts the foundation of authenticity from trust in preservation to confidence in reconstruction. If a claim is true, it can be verified again — independently, from what remains.
A deterministic approach treats proof as reconstructable rather than preservable: independent of attachment, cooperation, or metadata. We treat it as a design constraint. A system that claims to prove origin must continue to do so under the normal stress of real use: compression, editing, resharing, archiving, and time. If it cannot survive those conditions, it doesn’t solve the problem; it only moves it downstream.
The same principle applies broadly: proof must remain measurable and deterministic, independent of any platform’s goodwill. That’s not a philosophical stance — it’s a practical necessity for a connected world.
The story of C2PA stripping is a reminder that technological optimism is not enough. Good intentions and standards cannot overcome physics or network behavior. To make truth persist, we must design for loss, not for perfection.
We build the tools we build not to compete with standards, but to ensure that verification remains possible even when the system around it fails. A proof that disappears on upload was never proof — only a note of hope.
True authenticity lives in what can be verified again.
