The policy debate around AI-generated content keeps converging on labelling: watermark the synthetic, disclose the generated, badge the artificial. C2PA credentials are spreading through cameras and creative tools, and the AI Act's transparency rules push the same direction. All of it is useful. None of it is sufficient, because every labelling scheme shares one failure mode: the absence of a label proves nothing.
Provenance must attach to the authentic
Adversaries do not helpfully watermark their forgeries. Strip metadata, re-encode a file, screenshot an image, and most provenance signals vanish. The reliable direction is the inverse: cryptographically sign the authentic at the moment of capture or publication, so that anything unsigned is untrusted by default. Provenance becomes a property of real content rather than a confession attached to fake content.
Qualified seals make it legal-grade
This is precisely what qualified electronic seals do, with EU legal weight behind them. WeVerify's Qualified e-Seal binds an organisation's identity to content, documents, video, sensor data, AI outputs, with tamper-evidence and QTSP backing, €0.01 per seal. Newsrooms can seal originals, enterprises can seal records, and courts receive evidence whose integrity is checkable rather than arguable. The deepfake era does not need better labels for lies; it needs signatures on the truth.