Identity in the synthetic era

For most of history, identity verification worked because forgery was expensive. A convincing fake passport required craft, equipment and risk. A convincing fake face required a twin. The systems we built — show a document, compare a photo, trust what you see — all assumed that producing believable evidence of a person who does not exist, or impersonating one who does, was hard.

That assumption has quietly expired. Generative models can produce photorealistic faces of people who have never lived, clone a voice from a short sample, and render document images with plausible fonts, layouts and security-feature lookalikes. None of this requires specialist skill any more. The cost of producing believable fake identity evidence has collapsed, and it is not coming back.

This is what we mean by the synthetic era: a period in which the marginal cost of fabricating audiovisual "proof" of identity approaches zero, while the human ability to detect that fabrication stays flat. Every verification system designed for the pre-synthetic world now has to answer a new question — not "does this look right?" but "can this be proven right?"

It helps to be precise about which links in the verification chain fail. A photograph of a face no longer proves a person exists. A video of a person speaking no longer proves they said those words. An image of a driver licence no longer proves the licence was ever issued. What all three have in common is that they are media artefacts — recordings that were once expensive to fake and are now cheap.

What does not break is cryptography and physics. A digital signature over a credential still proves the credential was issued by the holder of a signing key. A liveness challenge performed in real time still requires a live subject to respond within human reaction times. A hardware-backed biometric sensor still measures a physical body part inside tamper-resistant silicon. The synthetic era does not end verification — it ends verification by inspection of media.

The Identity Wallet is built on exactly these three responses. The capture flow pairs document scanning with active liveness challenges, higher verification levels add hardware-backed biometrics and device integrity attestation, and everything a service receives is a cryptographically verifiable result rather than an image to eyeball.

• Challenge–response instead of static media. Active liveness — blink now, turn your head now — forces the evidence to be produced live, in response to an unpredictable prompt. Pre-rendered fakes cannot answer a challenge they did not know was coming.
• Hardware attestation instead of trust in software. When a fingerprint is read inside a phone's secure hardware and the device proves its own integrity (Google Play Integrity, Apple App Attest), an attacker needs to defeat physical silicon, not just generate pixels.
• Cryptographic provenance instead of visual inspection. A verified credential carries a signature chain back to the verification event. A service checking the signature does not care what the credential looks like — only whether the maths holds.

There is a tempting wrong answer to the synthetic era: collect more. If fakes are getting better, the argument runs, then verification should demand more documents, more biometrics, more behavioural data, held centrally so it can be cross-checked. This trades one failure mode for a worse one. Central stores of identity evidence are the highest-value target on the internet, and every record added makes the eventual breach more damaging — the honeypot problem.

The alternative is to make verification stronger while making disclosure smaller. Prove liveness without storing the video. Derive a verification result and discard the raw evidence from circulation — in the wallet's case, evidence is encrypted on the device before it moves anywhere, and services receive answers, never documents. A system that never accumulates a honeypot cannot leak one.

In the synthetic era, your face and voice are no longer secrets — assume they can be copied. What can remain yours is the ability to prove that a claim about you is true: that a live, verified person consented to this specific interaction at this specific time. That ability rests on keys you control, biometric checks performed against your physical presence, and consent records you can audit.

That is a genuinely different foundation from "a photo of my licence", and it is the foundation this service is built to provide. If you want to see what that looks like in practice, start with how it works or the guide to what each credential proves — and what it never reveals.