One touch. No squinting at traffic lights. It binds to your device's secure enclave — so a script, a headless browser, or an AI agent like Claude Code physically cannot complete it.
Press and hold the circle. Keep your finger or cursor on it until the ring fills.
Runs entirely in your browser. Nothing is uploaded. The biometric never leaves your device — only a signed yes/no receipt does.
Captchas fail because every puzzle an AI can see, an AI can solve. So we don't use a puzzle. We use signals a piece of software literally cannot produce.
A real hand never holds perfectly still. We sample micro-tremor, velocity variance, and timing jitter during the hold. A scripted press is too smooth and too regular — it scores near zero.
The browser stamps isTrusted on events that came from real hardware. A bot firing synthetic events with dispatchEvent cannot forge it — those events arrive isTrusted: false and we reject them.
The decisive step asks your device's secure enclave to assert presence with a real fingerprint, Face ID, or device PIN. This prompt is owned by the operating system, not the page. An AI agent, a headless browser, or remote code cannot satisfy it — there is no API to fake a fingerprint into a real authenticator. That's the difference between "hard for bots" and "impossible for software."
| Image captcha | This test | |
|---|---|---|
| Effort for a real person | Read, squint, retry | One touch |
| Solvable by modern AI | Yes — routinely | No |
| Solvable by a paid solving farm | Yes | No — needs the enrolled device |
| Accessibility | Poor (visual/audio) | PIN fallback, no vision needed |
| Proof you can keep | None | Signed, offline-verifiable receipt |
| Stores your biometric | — | Never leaves the device |
Straight talk: Layers 1–2 raise the cost for casual bots but a determined automation driver (e.g. a debugged browser) can imitate them. Layer 3 — hardware user-verification bound to an enrolled device — is the part with no software bypass. That is why the verdict below treats WebAuthn as the anchor and the behavioral signals as supporting evidence.