Proof of Human A Novel Identity Primitive for Verifying Unique Humans Online

Overview

Proof of Human is a novel identity primitive designed to verify a user as a real and unique human on the internet without revealing their personal identity. This capability addresses the increasing difficulty of distinguishing humans from advanced AI agents and automated bots, which often exploit traditional authentication methods.

Key Insights

• Problem Shift: Traditional defenses (IP limits, CAPTCHA, phone verification, device fingerprinting) and authentication systems (SSO, passkeys) fail because they rely on easily replicable proxies for a person or device. They solve one-to-one matching (is this user who they claim to be?), but not one-to-many matching (is this user unique among all others?).
• Uniqueness at Scale: Verifying a user as unique across billions of potential candidates requires biometric methods with extremely low error rates (e.g., 1 in 100 billion false matches).
• Privacy-Preserving Uniqueness: A robust system must verify uniqueness without ever identifying the user or storing their raw biometric data.
• Durable Credentials: Proof of Human credentials must persist across device loss and app reinstalls, anchoring to an abstract “account” rather than a single device.
• Agent Delegation: The system must support human users delegating actions to AI agents without treating the agents as adversarial bots.

Technical Details

A functional Proof-of-Human system requires five core pillars: Uniqueness, Anonymity, Recovery, Verification, and Delegation.

Uniqueness: One-to-Many Matching

Traditional authentication systems, like Face ID, perform one-to-one matching, comparing a single biometric capture against a stored template. This is mathematically simple, with a small comparison space.

Proof of Human, however, addresses a one-to-many matching problem. It verifies a user as unique against a potentially global population.

• Scale Challenge: A per-comparison error rate of one in a million, acceptable for one-to-one, becomes unviable at internet scale (e.g., 1,000 false matches against a billion candidates). Internet-scale uniqueness demands error rates of one in a hundred billion or better.
• Biometric Requirements: Most consumer biometrics lack the entropy for this scale. Highly distinct features like the iris pattern are crucial for achieving the necessary uniqueness guarantee.

Anonymity: Private Biometric Verification

Achieving internet-scale uniqueness without identifying the user presents a paradox. The solution involves a two-step approach:

Biometric Signal Acquisition

• High-Entropy Biometrics: The iris pattern offers the high entropy required for billion-scale comparisons, ensuring minimal chance of false matches between unrelated individuals.
• Secure Hardware Capture: Standard cameras are susceptible to spoofing. Purpose-built hardware, such as the Orb, is essential.
  • Orb Capabilities: Uses multispectral imaging (infrared and visible wavelengths), runs local neural networks for liveness detection and mask identification, and deletes original images before transmission.
  • Data Output: Transmits only signed, encrypted material derived from the iris, not the raw iris image itself.

Duplicate Checking via Secure Multi-Party Computation (SMPC)

The system checks for duplicates globally without any single party, including the system operator, ever seeing the raw biometric data.

• Process:
  1. The encrypted iris reading is split into three statistically random noise pieces.
  2. Each piece is distributed to a different organization (an SMPC node) in a distinct legal jurisdiction.
  3. Individual SMPC nodes cannot deduce the original reading from their piece.
  4. These nodes jointly compute comparisons (e.g., “does this reading match any existing enrollment?”) without ever reconstructing the full reading on any physical machine.
• Anonymized Multi-Party Computation (AMPC): This cryptographic technique ensures the user is verified as unique without identification.

Recovery: Durable Credentials for Lifelong Uniqueness

Proof of Human credentials must be recoverable after common events like lost phones or app reinstalls to serve as a lifelong property.

• Account-Based Credentials: Instead of a single secret, the system treats the verified human as an abstract account in a public registry.
  • The account stores a list of public keys belonging to Authenticators (software or hardware holding private keys, such as a wallet app, browser extension, or hardware security token).
  • The account itself does not store biometric data or user secrets.
• Verification Flow: Authenticators produce cryptographic proofs signed with their private keys. The public registry holds the matching public key for verification.
• WorldIDRegistry: An on-chain public registry stores one entry per verified human, detailing authorized Authenticators, designated Recovery Agents, and associated rules.
• Recovery Agents: In case of lost Authenticators, designated Recovery Agents (e.g., the Orb network) re-authenticate the human (typically via a fresh biometric check) and authorize the registration of new keys. Governance of these agents is an ongoing concern.

Verification: Privacy-Preserving Interaction with Services

Relying parties (e.g., online retailers, dating apps) require assurance of uniqueness while minimizing user data exposure.

• Relying Party Requirements:
  • Assurance of a real, unique human.
  • Ability to enforce one-per-human rules within their context (detecting repeat visits).
  • Confidence that only the fact of being a unique human is revealed, maintaining privacy of identity and activity.
• Nullifiers for Contextual Uniqueness: A nullifier is a number generated from three combined inputs: the user’s verified credentials, the relying party’s identifier, and a specific action.
  • The same input combination always yields the same nullifier. Different combinations produce uncorrelated nullifiers.
  • Relying parties can detect repeated actions by the same unique human if they see the same nullifier again, without knowing the human’s identity.
• Distributed Nullifier Generation: Nullifiers are not generated by the user or a single server due to trust and privacy concerns.
  • A distributed network of OPRF nodes (oblivious pseudorandom function) performs blinded computation on user queries.
  • The user unblinds the result, and a threshold of OPRF nodes must cooperate.
• Oblivious Nullifier Pool: An on-chain list of all used nullifiers prevents their reuse within a relying party’s context, mitigating pseudonymous profiling.
• Application Integration: The system leverages zero-knowledge proofs presented by the user to the relying party. The IDKit SDK streamlines this integration for developers.

Delegation: Enabling Human-Backed AI Agents

As AI agents increasingly act on behalf of humans, Proof of Human must accommodate this without treating beneficial agents as bots.

• Agent Registration: An AI agent registers its wallet address with a registry, like AgentBook (deployed on World Chain), linking it to a human’s World ID via a verification flow in the user’s wallet app.
• Service Verification: When an agent interacts with a service, the service can verify the agent is backed by a unique human, without identifying the human.
• Quota Enforcement: Agent activity counts against the human’s quota, preventing a single human from orchestrating many parallel agents for amplified reach.
• AgentKit SDK: Services integrate the AgentKit SDK to admit human-backed agents, often with default usage caps (e.g., three uses per human per service before payment).
• Use Cases: Enables bot-resistant free trials, fair limited-edition product drops, and enhanced fraud prevention for transactions.

Challenges and Future Outlook

The practical implementation of Proof of Human at internet scale faces several critical challenges:

• Hardware Decentralization: The long-term vision requires multiple independent manufacturers of secure capture devices (like the Orb), distributed across jurisdictions with cross-checking mechanisms. The architecture’s integrity relies on this transition.
• Unlinkability Under Adversarial Pressure: While the cryptography is sound in theory, its resilience against well-resourced adversaries (e.g., governments, large corporations) attempting to extract more data than intended needs continuous validation in practice.
• Delegation at Agent Scale: Current agent traffic is low. As autonomous AI agents proliferate, the per-human usage caps and the system’s ability to prevent new abuse patterns will undergo significant stress tests.
• The Bootstrap Problem: A Proof-of-Human system gains utility at scale, but achieving scale requires applications to demand it, which in turn requires users to adopt it. Navigating this adoption loop and managing the impact if a significant portion of the internet requires proof of human remains unclear.

The techniques underlying Proof of Human—one-to-many matching, secret-shared biometrics, scoped nullifiers, account-style key rotation, and delegated quotas—represent a deliberate effort to address identity problems that existing authentication infrastructure largely overlooks.