Answers

EU AI Act Article 50(2) Machine-Readable Marking — What Deployers Must Do

EU AI Act Article 50(2) requires providers of AI systems that generate synthetic audio, image, video, or text content to mark the output in a machine-readable format and label it as artificially generated or manipulated. The obligation is not optional disclosure -- it is a technical marking duty that binds from the moment the general-purpose AI transparency provisions became applicable in August 2026. For a verification-first operator, machine-readable marking is the natural bridge to a verified measurement credential: provenance metadata travels with the artifact, and an independent signature over the provenance record makes the marking auditable rather than self-asserted. The Council of AI treats Article 50(2) marking as a boundary-instrumentation problem, not a compliance footnote. Concretely: (1) every synthetic-output surface must emit a machine-readable marker carrying model identity, generation timestamp, and origin attestation. (2) That marker must be independently signed so a downstream auditor can verify 'who generated this' without trusting the generator's own word. (3) The marking chain feeds the Council of AI Containment Incident Index, so a mislabeled or unmarked output is traceable to a boundary event. The posture is monitored containment, not provable isolation -- marking does not prove content is harmless, it proves the provenance path is watched and verifiable.

Measurement, not certification. Verify a card at /gspc-verify.