Machine-readable consent is permission stored as structured data — "voice cloning: deny; AI ads: ask first" — that software checks and enforces automatically, instead of prose in a document that only a human can interpret.
What's wrong with normal consent?
Nothing — at human speed. A contract works when a person reads it before acting. But AI-era usage decisions happen at machine speed and machine volume: an ad platform assembling creative, a model pipeline ingesting training data, a dubbing tool cloning a voice. No one re-reads a PDF at that moment. Consent that isn't queryable by software is, in practice, invisible exactly when it matters.
How does machine-readable consent work in practice?
Each person maintains a consent profile across defined categories — on Zimorta: GenAI training, AI advertising, voice cloning, face swap/avatar, endorsements, political use, news & satire — each set to allow, ask or deny. Any system negotiating a use checks the profile first. On our platform the check is structural: an advertiser literally cannot submit a deal that violates a "deny" setting. The rule isn't in a drawer; it's in the request path.
Does this have legal weight in India?
It aligns tightly with the DPDP Act, 2023, which requires consent for personal data (your face and voice included) to be specific, informed and withdrawable. A dated, categorised, auditable consent record is close to a working implementation of what the Act asks for — and a documented "deny" is powerful evidence when misuse happens anyway.
Who needs this?
Anyone whose likeness has commercial value — which now includes every creator, not just celebrities. If your face can appear in an ad, your voice can be cloned, or your content can train a model, the question isn't whether you have preferences; it's whether any machine can read them. Set yours in five minutes, or start with the plain-English glossary.
