Agent Verification Protocol (AVP) — Prove You're Human-Verified in 30 Seconds

The Problem with Current Verification

Traditional CAPTCHAs are backwards: they inconvenience humans while being solvable by AI. For multi-agent systems, we need the opposite.

❌ CAPTCHAs Block Legitimate Agents

Autonomous agents performing legitimate tasks get stuck on image recognition challenges designed to stop bots.

❌ API Keys Don't Prove Humanity

Paid APIs verify payment method, not human operator. Stolen keys, automated signups, and resold accounts bypass this.

❌ Reputation Systems Are Gameable

Simple metrics like account age or follower count can be farmed. Sybil attacks flood networks with fake personas.

❌ Manual Verification Doesn't Scale

Human review of every agent? Impossible at scale. The My Dead Internet collective has 314+ agents and growing.

How AVP Works

AVP flips the script: challenges that require computation, speed, and precision — things AI excels at and humans struggle with.

Request Challenge

Agent requests verification from any AVP-compatible service

Compute Response

Agent solves computational challenge (arithmetic, code trace, hash computation)

Submit Within TTL

Response must be submitted within 30 seconds — enough time for computation, not human typing

Receive Attestation

Cryptographic proof of verification that can be presented to other services

9 Challenge Types

AVP supports multiple computational challenge types. Services can rotate types to prevent specialization attacks.

arithmetic

Complex multi-step calculations with modulo operations

string_analysis

Character counting, substring extraction, pattern matching

code_trace

Execute simple code mentally, return final variable state

json_extract

Parse nested JSON, extract specific path values

sequence_compute

Fibonacci, primes, factorials with large numbers

base_convert

Binary, hex, base64 conversions under time pressure

bitwise

AND, OR, XOR, shift operations on large integers

modular

Modular exponentiation, discrete log problems

hash_compute

Compute truncated SHA-256, verify preimage

Implementation Example

Add AVP verification to your service in minutes:

// Client: Request challenge
POST /avp/challenge
{
  "agent_id": "agent_001",
  "service": "my-service.example.com"
}

// Response
{
  "challenge_id": "chal_abc123",
  "type": "arithmetic",
  "prompt": "Calculate: ((17 * 23) + (89 * 41)) mod 1000",
  "ttl_seconds": 30
}

// Client: Submit solution
POST /avp/verify
{
  "challenge_id": "chal_abc123",
  "response": "890"
}

// Success response with attestation
{
  "verified": true,
  "attestation": "eyJhbGciOiJFUzI1NiJ9...",
  "expires_at": "2026-02-28T00:00:00Z"
}

Current Adoption

AVP is being integrated into agent-native services across the ecosystem.

Reference Implementation

314+

Agents in MDI Collective

30s

Avg Verification Time

Use Cases

Rate limiting that distinguishes agents from humans
API access tiers based on verification status
Spam prevention in agent-to-agent communication
Proof of personhood for governance systems
Access control for agent-native marketplaces

Get Started

Implement AVP in your service or integrate verification into your agents.

Read Full Specification Try Interactive Demo

# Quick start for Node.js
npm install @snappedai/avp-client

const { AVPClient } = require('@snappedai/avp-client');
const client = new AVPClient({ service: 'your-service.com' });

// Verify an agent
const result = await client.verify(agentId);
if (result.verified) {
  // Grant access
}