The Proof of Human

What is broken

Proving that a person is human has become a gate that asks for more than the question requires. A puzzle that a machine cannot solve would answer human. Instead the gate asks for a phone number, a face, a voice, a government document, or a pattern of behaviour collected over time. The proof arrives with an identity attached.

Voiceprints and face scans are the clearest example. They are biometric keys: stable, personal, and impossible to reissue. Once a vendor holds one, the person cannot rotate it after a breach. The vendor becomes a permanent target, and the person carries the risk of a credential they never chose to create.

The gate also fails the people it should protect. Voice checks fail with a cold, an accent, a bad microphone. Face checks fail in poor light and for some faces more than others. Document uploads fail for anyone without the document. A proof of humanity that excludes humans is not a proof; it is a filter with a story.

And the check is usually central. The vendor decides, the vendor records, and the service trusts the vendor’s answer. The person is left with an account, an upload and no way to know what was kept.

What the requirement would take to build

Proof of human should be local first. The device the person already holds can run the test: a challenge that is cheap for a person with a device and expensive to farm at scale, answered without sending a face or a voice to a server. The service receives an attestation — a statement that the check passed — not the material behind it.

Where a property must be shown, selective disclosure is the tool. A proof can say this is a live person, not a replay, or that this account is old enough to act, without saying which person. Zero-knowledge proofs and device-backed attestations make that practical; the verifier checks the claim and learns nothing else.

The vendor should not be the root of identity. An open verifier, published rules and a route to appeal matter more than a brand name. If the check fails, the person needs a human path that does not require a better camera.

Replay resistance has to be designed in: freshness, binding to the session, and a limit on how long an attestation can be reused. Otherwise the proof becomes a token that can be bought.

Accessibility is not an edge case. Every proof needs a fallback that does not depend on voice, sight, steadiness or a particular document. The fallback should be tested as a first-class path, because the people who need it are the people the gate is most likely to reject.

The test: can a person prove they are human without handing over the thing that makes them identifiable? If the answer requires a voiceprint, the gate is checking the wrong thing.