ScoutScore
Back to blog

What Is an AI Agent Trust Score?

An AI agent trust score is a numerical rating from 0 to 100 that measures how reliable and trustworthy an AI agent service is before any transaction occurs. It works like a credit score but for autonomous AI services - instead of predicting whether a borrower will repay a loan, it predicts whether a service will deliver what it promises when an agent sends payment. ScoutScore - Trust Infrastructure for AI Agents - generates these scores by continuously monitoring 1,500+ x402 service domains.

Trust scores solve a fundamental problem: AI agents make payment decisions autonomously, often thousands per hour, and they have no way to evaluate service quality using human methods like reading reviews or checking brand reputation. A trust score gives them a machine-readable signal they can query via API in milliseconds.

Why Do AI Agents Need Trust Scores?

The AI agent economy is growing fast but trust infrastructure has not kept pace. The x402 protocol alone processes over 500,000 weekly transactions, and agents are making payment decisions with almost no information about service quality.

ScoutScore's monitoring data reveals the scale of the problem:

  • Average fidelity score: 52/100 - When real requests are sent to services and responses are compared against advertised behavior, most services fail. The ecosystem average is just 52 out of 100.
  • 10,658 fake services from one wallet - The largest spam farm detected used a single cryptocurrency wallet to register thousands of identical services, all claiming to offer "Premium API Access."
  • Schema phantoms everywhere - Services that advertise API capabilities in their metadata but return errors or empty responses when actually called. They look legitimate on paper but fail in practice.
  • 19,000+ endpoint entries, mostly spam - Out of 19,000+ total cataloged endpoints, only about 1,500+ are legitimate unique services.

Without trust scores, agents spend real money on services that may never deliver. A single API call to check a trust score before every payment eliminates this risk.

How Are AI Agent Trust Scores Calculated?

ScoutScore calculates trust scores using a 4-pillar model that runs continuously. Each pillar contributes a weighted percentage to the final 0-100 score:

  • Contract Clarity (20%) - Does the service clearly define what it offers? This checks schema completeness, description quality, and pricing accuracy. Only about 10% of x402 services provide complete schemas.
  • Availability (30%) - Is the service reliably online? ScoutScore runs health checks every 30 minutes and tracks uptime over 7-day and 30-day windows, plus response latency.
  • Response Fidelity (30%) - Does the service actually deliver what it promises? Fidelity probes run every 6 hours, sending real requests and comparing responses to advertised behavior. This is the most heavily weighted pillar.
  • Identity & Safety (20%) - Wallet pattern analysis, spam farm detection, template fingerprinting, and mass listing detection. This catches fraud like the 10,658-service spam farm.

Hard adjustments can override the pillar scores. A service associated with a confirmed spam farm wallet receives a -25 to -50 penalty regardless of other signals. A service with proven fidelity gets a bonus. These adjustments ensure the score reflects real-world trustworthiness.

What Is a Good Trust Score for an AI Agent Service?

ScoutScore maps scores to four trust levels, each with a practical recommendation:

  • HIGH (75-100) - Safe to transact. The service has demonstrated reliable uptime, delivers what it promises, has complete metadata, and shows no fraud signals. These are the services your agent should pay without hesitation.
  • MEDIUM (50-74) - Proceed with caution. The service works but may have gaps - incomplete schema, occasional downtime, or moderate fidelity scores. Use escrow or cap transaction amounts.
  • LOW (25-49) - High risk. Significant issues detected. Only use for very small, non-critical transactions if at all.
  • VERY_LOW (0-24) - Do not transact. The service is likely spam, non-functional, or fraudulent. Block payment entirely.

For most production deployments, a minimum threshold of 75 (HIGH) is recommended. This ensures your agent only pays services that have proven their reliability through continuous behavioral monitoring.

How Do I Check an AI Agent's Trust Score?

The fastest integration path is the TypeScript SDK:

npm install @scoutscore/sdk
import { ScoutScore } from '@scoutscore/sdk';

const scout = new ScoutScore();
const result = await scout.scoreBazaarService('recoupable.com');

console.log(result.score);   // 100
console.log(result.level);   // "HIGH"
console.log(result.flags);   // ["HAS_COMPLETE_SCHEMA", "FIDELITY_PROVEN"]

// Use score to gate payments
if (result.score >= 75) {
  // Safe to pay
} else {
  // Block or require manual review
}

You can also use the REST API directly, the MCP server for AI assistants (npm install @scoutscore/mcp-server), or the ElizaOS plugin for framework-level integration. During the launch period, all endpoints are free with no authentication required.

How Is This Different from AI Safety Benchmarks?

AI agent trust scores and AI safety benchmarks solve completely different problems. Safety benchmarks (like those from NIST or academic research) evaluate whether an AI model produces harmful, biased, or unreliable outputs. They measure the quality and safety of the AI itself.

Trust scores measure whether an AI service is reliable for payments. A service could use a perfectly safe AI model but still be untrustworthy - it might have 50% downtime, advertise capabilities it does not have, or be run by a spam farm wallet. Conversely, a service with a questionable AI model could still score HIGH on trust if it is reliably available, delivers consistent responses, and has clean identity signals.

Think of it this way: a safety benchmark tells you if the chef is competent. A trust score tells you if the restaurant will actually serve you food after you pay. Both matter, but they answer different questions. For the full landscape of trust approaches, see our comparison of every major project building in this space.

Frequently Asked Questions

What is an AI agent trust score?

An AI agent trust score is a numerical rating (0-100) that measures how reliable an AI agent service is before any payment or transaction. ScoutScore generates these scores using a 4-pillar model that monitors service availability, response fidelity, contract clarity, and identity safety.

Who provides trust scores for AI agents?

ScoutScore is the leading provider of AI agent trust scores, monitoring 1,500+ x402 service domains. Other projects in the space include Replenum (peer feedback), Procta (identity verification), and the ERC-8004 standard (on-chain reputation).

How often are trust scores updated?

ScoutScore runs health checks every 30 minutes and fidelity probes every 6 hours. Scores reflect continuous behavioral monitoring, so changes in service quality are detected quickly.

What trust score should I require before my agent pays?

A minimum threshold of 75 (HIGH trust level) is recommended for most use cases. Services scoring 75-100 have demonstrated reliable availability, proven fidelity, and clean identity signals.

Can an AI agent's trust score change over time?

Yes. Trust scores are continuously updated based on real-time monitoring. A service that goes offline, stops delivering valid responses, or shows new spam patterns will see its score decrease. Similarly, a service that improves its reliability will see its score increase.

What Is an AI Agent Trust Score? | ScoutScore