EU AI Act · Article 50

EU AI Act Article 50 and marking AI-generated content

Marking obligations came into application in August 2026. The technical answer is broadly settled. The evidentiary answer, showing that you marked what you were required to mark, is not.

Article 50 of the EU AI Act sets transparency obligations for certain AI systems. Providers of systems generating synthetic audio, image, video or text must mark outputs in a machine-readable format detectable as artificially generated or manipulated. Deployers of systems producing deep fakes have disclosure obligations of their own.

The technical mechanism most of the industry has converged on is content provenance: cryptographically signed manifests bound to the asset, carrying a declaration of how it was made.

What marking does and does not settle

Marking answers the question asked of the asset: is this artificially generated? It does not answer the question that eventually gets asked of the provider: can you demonstrate that your systems marked what they were supposed to mark, over a period you no longer control?

Those are different obligations with different evidence requirements. The first is satisfied by the asset carrying a credential. The second requires a record of what the system did, retained and verifiable.

Where marking is fragile

  • Manifests can be stripped. Platforms re-encode, pipelines rewrite, and a stripped asset is indistinguishable from one that was never marked.
  • Absence proves nothing. If a credential is missing, no one can tell whether the provider failed to mark it or a distributor removed it.
  • Records are self-produced. A provider's internal log that marking occurred is produced by the provider.

Anchoring closes the gap

Recording the manifest hash to a public ledger at the moment of generation creates evidence that survives stripping. The credential may be removed downstream, but the fact that a specific asset was credentialed at a specific time remains provable from infrastructure the provider does not control.

The compliance sentence becomes concrete: Article 50 makes you mark it. An anchor proves you did.

Check it yourself

Every Rubric attestation resolves publicly, with no account and no API key, and every anchor resolves to a public ledger message you can read without our cooperation.

HCS topic 0.0.10416909 · ML-DSA-65 signatures

Verify an attestation · Read the ledger ↗

Standards alignment

Rubric's evidence documents carry C2PA Content Credentials, the specification the major platforms have adopted, layered over post-quantum attestations anchored to a public ledger.

Verifiable, not asserted

Rubric Protocol is a conformant C2PA Generator Product (Content Credentials 2.4, Assurance Level 1).

Record 01a002b7-3663-7b3b-a60e-db3b99ee2d94 · Echelon Intelligence Group LLC

Primary-source records

Related: C2PA for documents · Annex IV technical documentation · Trust & standards