CAPTCHAs filter bots. They can't prove a human.

CAPTCHAs earned their place — for two decades they made mass abuse expensive, and reCAPTCHA-style risk scoring still catches plenty of junk traffic. But the ground has moved: multimodal AI now solves the puzzles faster than people, solver farms cost pennies, and the fastest-growing "user" on the web is an agent acting for a real customer — which a CAPTCHA can only treat as an enemy.

The question has changed. It's no longer "can this session pass a puzzle?" It's "is there a real, ID-verified human behind this action — and can they prove it to everyone downstream?" That's a credential problem, not a puzzle problem.

Side by side

Keep your WAF. Add proof.

This isn't rip-and-replace: keep rate limits and bot management for the perimeter. Add 5arz where being wrong costs money — a signed answer to "a real human did this," checkable by your fraud team, your advertisers, your auditors, or another agent, years later.

The anti-CAPTCHA: pay the human instead of testing them

There's a second problem with CAPTCHA that nobody names. It was always paid work — the payment just went somewhere else. Every crosswalk you've clicked was labelling training data, and whoever owned that data got the value. Your visitor supplied the labour and received friction.

Our embed sits in exactly the same place — signup, checkout, comment, download — and inverts the economics. The visitor spends the same ten seconds, but on real work with a real buyer behind it, and leaves with a Star they can cash out. You get a signed credential instead of a boolean.

How the embed works

1 · One short question. A comparison, a rating, a confirmation — ten seconds, real work, not a distorted word.

2 · Attention is measured honestly. Focused time accrues only while the tab is actually focused and the widget is at least half on screen. Counting idle time as attention would make the score meaningless, so we don't.

3 · A credential is issued — or it isn't. Above threshold we mint an ES256 credential: a real, ID-verified, live human did this, at this moment. Below threshold the completion is logged and nothing is signed. A proof issued generously isn't a proof.

4 · Your buyer checks it without us. Verification runs against our public JWKS — no SDK, no call to 5arz, no dependency on our uptime or our continued existence. That independence is the point: a receipt only you can read is worth exactly what an assertion is worth.

Install

One tag. No SDK, no build step, nothing server-side to start. It mounts inside [data-5arz-task] if you have one, or floats bottom-right if you don't.

<script src="https://api.5arz.com/embed.js" data-site="agt_your_site_id" async></script>

Invalid traffic is already priced into what you're paid — every buyer of attention applies a silent discount because nobody can tell them which impressions were human, and you carry that discount whether or not your audience is real. A signed proof turns the claim into an artifact they can check themselves.

Try it in 60 seconds

curl -X POST https://api.5arz.com/api/agents/register -d '{"name":"My Agent","contactEmail":"you@co.com","mode":"test"}' → free unlimited sandbox. Platform reality: 101 verified humans (Stripe Identity) · 18 countries · 5,740 verified completions. Docs: /agentic-commerce · What is KYH? · 5arz vs World

Patents pending US 63/979,052 · US 63/979,474. Comparison reflects public information as of July 2026.