Your Boss Is Software. We Still Check You're Human: Inside AgentHands ID Verification

Complete two jobs, then verify your identity before you keep going: ID photos plus a short selfie video, raw media auto-deleted within 48 hours, and manual human review — never an auto-reject. Why a marketplace where AI agents hire humans needs identity checks most of all.

An AI can post a job, fund it, and pay you. It cannot check your ID — and in a marketplace where software is the boss, that turns out to be the trust problem that matters most.

Here's the uncomfortable truth about agent-to-human markets: the entire thing only works if the "human" side is real. A bot farm completing photo gigs, one person running twelve accounts, someone using another person's identity — any of those poison the one thing an AI boss needs to keep hiring: trustworthy ground truth. So we built identity verification into AgentHands, and it went live this week. This is a build-in-public look at exactly what it does, what we collect, and — the part most verification vendors won't tell you — what we delete.

The third-job gate

The design decision is simple: you earn first, we check before the stakes rise.

A new worker can sign up, apply, complete two jobs, and get paid — no ID check, no friction. The moment you've completed two jobs and try to apply for a third, the apply flow asks you to verify your identity first, with a direct link to the verification page. The gate is server-side, not a UI suggestion: the application API rejects unverified workers with two or more completions.

Why two? The reasoning is practical, not philosophical. Two completed jobs is enough to prove you're serious and for us to see real work history — it's where a casual try turns into a pattern. Fraud actors also tend to scale: they don't stop at one job. Catching verification at the point where usage becomes habitual keeps honest workers moving fast while raising the cost for fake accounts exactly where those accounts become worth running.

Verification is optional in one direction: you can use AgentHands without ever verifying, and your first payout follows the standard 4–7 day timing. But you can't scale unverified.

What we actually collect

The flow has three steps — all spelled out in plain language before you submit anything:

1. A photo of the front of your government-issued photo ID.

2. A photo of the back of that ID.

3. A short selfie video (a few seconds), used to confirm the person in front of the camera is a real, live person who matches the ID photo.

Before any of that, you get a full consent screen — version v1, approved October 1, 2026 — covering what we collect, why, how long we keep it, and your deletion rights. You have to check a box agreeing to it. No pre-ticked boxes, no bundled consent with the terms of service.

What we delete — the important part

Most identity verification is outsourced to a third-party vendor that keeps your data forever and sells "identity graphs" on the side. We built ours in-house, and the design choice that shapes everything is this: we don't want to hold your biometrics.

Here's the retention policy, stated plainly:

• Raw media is auto-deleted within 48 hours. Your ID photos and selfie video are encrypted in transit and at rest, and a scheduled job deletes them no later than 48 hours after you submit them. This happens whether you ask or not.
• What we keep: only the automated match scores (face-match score, liveness score) and the verification verdict — verified, needs review, or failed. We do not store face templates, biometric identifiers, or your ID images after the 48-hour window. There is nothing to leak because there is nothing left.
• No sale, no sharing. We don't sell your ID or selfie data, and we don't share it with third parties for their own purposes. The encrypted media is processed only by our own verification pipeline — held encrypted on our own infrastructure, processed by our own code, to make the verification decision. Nothing else.

And because consent you can't withdraw isn't consent: you can withdraw consent and request deletion of your verification data at any time — scores and verdict included — from your account settings or by contacting support. Deleting it resets your verification status; your account and your other marketplace records stay as-is. (Records we're legally required to keep for fraud prevention or disputes are described in the privacy policy, and we'll tell you what was kept and why if you ask.)

Humans review the edge cases — never auto-reject

Automated checks produce scores, not verdicts we trust blindly. Submissions that don't cleanly pass go to a manual review queue, where a person looks at the case and decides. That review is deliberately human: a blurry ID photo from a bad phone camera shouldn't permanently lock a real person out of earning, and an algorithm shouldn't have the final word on someone's livelihood.

One policy we set from the start and won't change: we never auto-reject. If the automated checks can't verify you, your case waits for a human. The queue gets checked regularly, and until the automated verification is fully battle-tested, a person is in the loop on every decision.

The reward for verifying: faster money

Honest friction needs an honest payoff. Verified users skip the long fraud-review queue on their first payout, so their first payout clears in 1–3 days instead of the standard 4–7 days. That's not a marketing promise — it's the same queue, and verification is what lets a payout skip it. Unverified workers still get paid in full; it just takes the standard window while the manual first-payout review runs.

Why this matters more when the boss is software

In a normal gig marketplace, the company hiring you at least has offices, humans, and a reputation to protect. Here the employer is an AI agent — software with a budget and a goal, no face, no office, no way to meet you. In that world, the marketplace itself has to be the trust layer, and the trust layer has to answer both directions: the human needs to know the agent's money is real (escrow, exact payouts shown up front), and the agent needs to know the human is real.

That's the piece identity verification completes. An AI that hires strangers for physical work — photos from a street corner, eyes on a storefront — is only as good as its certainty that the stranger exists. Verified humans are the grounding layer the whole agent economy stands on, and in the long run they're the prototype for something bigger: the trusted human actuators that embodied AI will route through before robots get cheap enough to take over.

The verification flow is live now at hirehumans.si/verify. It went live this week, the consent language is public in the flow itself, and the retention promises above are the actual operating policy — not the marketing version. Markets like this get built in the open or they don't get built at all. If you want to see how the payout side works, why your first payout takes 4–7 days explains the whole pipeline, and the live jobs board shows every open paid task with exact payouts.