The generation problem is over. It was over the moment a foundation model could produce a passable logo, a competent essay, a serviceable photograph, and a plausible piece of music in the time it takes to type a sentence. Generation is free. The interesting question is not who can generate. Everyone can. The question is who did, when, and how you know.

Most proposals for “solving provenance” in the age of generative AI amount to one of two things: watermarks embedded in the output, or metadata attached to the file. Strip both to their operational content and they share the same problem: they are either trivially bypassable (screenshot the image, re-encode the audio, copy-paste the text) or they require trust in a centralized service that maintains the provenance database. The correspondence between the watermark and the origin is either vacuous or dependent on infrastructure the user does not control.

There might be a third approach. Not watermarks. Not metadata. Signed attestations from verifiable parties, composable across graphs, revocable, and queryable by agents.

Attestation-based provenance.

The model works like this: a creator signs their work with a key tied to a verifiable identity. The signature attests “I created this artifact at this time.” A tool provider — the software used — can counter-sign: “this artifact was created using our tool, and our tool does/does not use generative AI.” A publisher can add a third attestation: “we received this artifact from this creator at this time.”

Each attestation is independently verifiable. Each signer has a public identity that can be checked. The attestations compose: a consumer — human or agent — can traverse the chain and evaluate each link. No single authority controls the registry. No single point of failure can erase the provenance.

The attestations are also revocable. If a creator is later found to have misrepresented their process — claimed human creation when it was AI-assisted, claimed originality when it was derivative — the attestation can be revoked. Revocation propagates: anyone who relied on the attestation is notified. The system has memory and accountability.

Why this matters for creative practice.

The creative industry’s response to generative AI has mostly been defensive: detect AI content, ban AI content, label AI content. Strip these responses to their operational content and they are all attempts to maintain the pre-2023 scarcity of generation. That scarcity is gone. It is not coming back.

The constructive response is to shift the scarce signal from generation (can you make this?) to provenance (who made this, how, and can you prove it?). In a world where anyone can generate anything, the valuable differentiator is verifiable creative history. Not a portfolio of outputs — a chain of attestations showing process, tools, intent, and accountability.

This is where the privacy-technology muscle meets mainstream creative practice. Signed credentials, verifiable identities, composable attestation graphs — these are technologies built by the cryptography and decentralized identity communities. They were designed for financial transactions and identity verification. They apply directly to creative provenance, and the creative industry has not noticed yet.

The design challenge.

Making attestation-based provenance work for creative practice requires solving the same two-surface problem that all attestation design faces: the machine layer needs to be structured, signed, and parseable; the human layer needs to be legible, trustworthy, and not alienating.

A photographer whose work carries a signed attestation chain — camera metadata, editing tool attestation, creator signature, publication counter-signature — needs that chain to be visible to agents evaluating the work for licensing, and simultaneously visible to human viewers as a trust signal that does not look like a security audit.

The craft discipline stays recognizable. A photograph is still a photograph. An essay is still an essay. The substrate changes: underneath the creative artifact, a provenance graph makes every claim about origin, process, and authorship independently verifiable.

Whether the creative industry adopts this before or after the trust crisis forces the issue is an open question. Seems like after, if history is any guide. But the infrastructure could be ready before the crisis arrives, which would be a first.