ProofPass: verified-human credentials for agentic checkout

The problem

AI shopping agents now browse and buy like humans, and fraud systems cannot tell a legitimate agent acting for a verified human from a bot. Chargebacks911 warns that legitimate AI-initiated purchases are being misclassified as fraudulent bot activity, triggering false declines and revenue loss at scale, while the same systems face genuine agentic fraud on the other side. Per PYMNTS Intelligence, 56% of consumers would let an agent search and compare, but only 35% would hand it saved payment credentials: the trust layer is missing.

The idea

Why now

Three launches converged this week: World's World ID for banking and finance plus AgentKit (Oct 18, 2026, Money20/20) gives consumers a delegable proof-of-human for agents for the first time; vLLM v0.32.0 shipped POST /v1/systemone, putting typed choice/noul/score judgments inside the inference engine for cheap policy gates; and agentic commerce hit the mainstream - CHEQ reports AI agent visits to retail sites more than doubled year over year, with Visa and Mastercard piloting agent-initiated transactions.

What it combines

World ID AgentKit (world-id-banking-finance-identity-suite) supplies the identity primitive: a verified human delegating proofs to an agent. vLLM's /v1/systemone (vllm-v0-32-0-systemone-decision-endpoint) supplies typed policy judgments at checkout. Jev-style decision gating turns yes/no policy evaluation into a calibrated audit trail. The mix matters because identity proof without policy gates is a blunt instrument (accept/reject), and policy gates without cryptographic identity are just another fraud score. Together they answer the two questions a merchant actually asks: is this agent acting for a real person, and is this purchase within the authorization?

MVP

Weekend MVP: a Shopify checkout webhook that (1) verifies a World ID proof presented by an agent via AgentKit, (2) evaluates three hardcoded typed gates - spend cap, category allowlist, velocity limit - and (3) writes an audit log (proof hash, gate results, order id). Deliberately skip: multi-issuer support, non-Shopify platforms, dynamic policy authoring, dashboard UI.

How it makes money

The moat

The largest fraud incumbents (Riskified, Forter, Signifyd) monetize uncertainty: probabilistic scoring sold as SaaS. A cryptographic proof product cannibalizes their core - if identity is provable, scoring revenue shrinks, so they cannot ship it without undermining their own pricing. They also have no identity network: matching it means 18M+ verified humans and Orb-scale hardware World already deployed. This is business-model conflict plus a credential network, not an execution claim. Honest anti-moat: World itself is moving merchant-ward, so ProofPass must be identity-agnostic from day one; the moat is the multi-issuer verification layer, not World exclusivity.

Distribution

B2B2C via the Shopify App Store first: wedge into mid-market merchants already seeing agent traffic and false declines (CHEQ data shows retail attracting roughly 1 in 5 agent visits across its network). Fraud-stack partner integrations as the second wedge; the merchant's fraud team becomes the internal champion.

Why it wins

Riskified, Signifyd, Sift, and Forter sell probabilistic fraud scoring: they guess. ProofPass verifies: a cryptographic proof that the agent is delegated by a verified human, plus explicit typed policy gates. It solves the revenue-unlock problem (accept blocked good traffic) that fraud scorers structurally cannot, because a score can never prove a human authorized the purchase.

Risks

Biggest risk: credential supply is concentrated - roughly 18M World holders, skewed by Orb geography, so many merchants' customers cannot present proofs yet. The MVP de-risks it by targeting the narrow slice of merchants already reporting agent-traffic false declines, where even partial coverage converts blocked orders; the audit log also makes the coverage rate visible, so the go/no-go is data, not hope.

Build it with

• World ID for banking and financeThe identity primitive: a verified human delegating World ID proofs to an agent, launched Oct 18, 2026. Supplies the cryptographic credential ProofPass verifies at checkout.
• vLLM v0.32.0In-engine typed choice/noul/score judgments. The policy gate that evaluates spend cap, category allowlist, and velocity limit per agent order.
• autotrust JEV-35B: 35B-A3B open decision model with calibrated probabilitiesOpen decision model with calibrated probabilities: the Jev-style decision layer that turns gate evaluations into a calibrated, auditable approval record.

Repo to start from

proofpass-shopify — Shopify checkout webhook verifying World ID agent proofs with typed policy gates (spend cap, category allowlist, velocity) and per-order audit logs.

Evidence

• Sam Altman's World Debuts ID Verification for AI Agents (PYMNTS)
• World Pitches Banks a Bundle of Human-Verification Tools to Counter Deepfakes (Unchained)
• AI shopping agents are being wrongly flagged as fraud, costing merchants legitimate revenue (Chargebacks911)
• Agentic Commerce, Fraud and Chargebacks: Money20/20 Insights (Chargeflow)
• Kount vs Sift vs Forter: $160K Fraud-Tool Gap (2026)
• Black Friday AI Traffic: What Retailers Need to Know (CHEQ)