Introducing CAR — Every AI agent gets a license plate

Trust Scoring for
AI Agents

VORION provides the infrastructure to bind AI agents to verifiable human intent. Real-time trust scoring, capability gating, and immutable audit trails.

New to AI Governance?

What is AI Governance?

Think of it like a credit score for AI. Just as banks use credit scores to decide loan amounts, AI governance uses trust scores to decide what an AI agent can do.

1

AI Takes Action

An AI agent wants to do something—send an email, make a purchase, access data. Every action has a risk level.

2

Trust is Checked

The system checks: "Has this AI earned enough trust for this action?" New agents start with low trust and must prove themselves.

3

Decision is Made

High trust? Approved. Low trust? Denied or escalated to a human. Everything is logged for compliance.

The 8 Trust Tiers (T0-T7)

T0
Sandbox
0-199
Isolated testing only
T1
Observed
200-349
Read-only, human review
T2
Provisional
350-499
Basic writes, policy-checked
T3
Monitored
500-649
Standard ops, anomaly detection
T4
Standard
650-799
Extended ops, reduced review
T5
Trusted
800-875
Privileged ops, express path
T6
Certified
876-950
High autonomy, council review
T7
Autonomous
951-1000
Full autonomy, auto-approve

Trust is earned through consistent good behavior and lost quickly through failures. Every agent's trust history is bound to their unique CAR ID.

The Challenge

AI agents are deployed without governance infrastructure, creating compliance and security risks.

Ungoverned Agents

AI agents operate without trust boundaries, making unrestricted decisions with no accountability.

No Audit Trail

Enterprises can't prove what AI did, when, or why—a compliance nightmare.

Trust Vacuum

No standard way to measure, verify, or communicate how much trust an AI agent has earned.

Governance Platform

Built on the BASIS open standard, implemented via the CAR client (@vorionsys/car-client) — TypeScript types and contracts you install from npm.

Trust Scoring

0-1000 credit-score model with 8 discrete tiers (T0-T7). Weighted across observability, capability, behavior, and context dimensions.

T7: AutonomousT5: TrustedT4: Standard

Capability Gating

Every action is checked against trust level. Insufficient trust? Request denied, escalated, or degraded automatically.

ALLOWDENYESCALATEDEGRADE

Proof Chain

SHA-256 hashed audit trail with cryptographic verification. Every decision is provable, every action is traceable.

prf_a7b2c9d4e5f6...

Trust Decay

Trust isn't permanent. Scores decay over time, with 3x accelerated decay after failures. Continuous good behavior required.

7-day half-life | 3x failure multiplier | Real-time updates

See It In Action

Watch how Vorion governance evaluates and governs AI agent actions in real-time.

Aurais
Trust ScoreT3 Monitored
687/ 1000

The Cohesive Stack

Four layers, one mission: safe autonomous AI. Register, parse, check, log, anchor.

The Law

BASIS

Open standard for AI governance. The rules every agent must follow.

The DMV

CAR

Categorical Agentic Registry. Every agent gets a unique CAR ID — trust scores, capabilities, and audit history bound to one identity.

The Engine

Cognigate

Enforcement runtime for AI actions. Validates against policies and gates execution in real-time.

The Receipts

PROOF

Immutable audit trail for AI decisions. Every action is logged, hashed, and anchored for compliance.

Processing Pipeline
REGISTER (CAR)PARSE (INTENT)CHECK (ENFORCE)LOG (PROOF)ANCHOR (CHAIN)

“BASIS sets the rules. CAR identifies the agent. Cognigate enforces the decisions. PROOF keeps the receipts.”

Works With Your Stack

Callback-based integration means no architectural changes required.

LangChainCrewAIAutoGenCustom Agents
import { TrustBand, TRUST_THRESHOLDS } from '@vorionsys/car-client';
import { createTrustEngine } from '@vorionsys/atsf-core';

const engine = createTrustEngine();
await engine.initializeEntity('agent-001', TrustBand.T2);

// Your existing code - unchanged
const callback = engine.createCallback('agent-001');
await agent.invoke(input, { callbacks: [callback] });

Get Started

Interested in AI governance for your organization? Let's talk.

Join the Community

Chat with developers building with Vorion.

Join Discord

Open Source

View source, contribute, or fork for your needs.

View on GitHub

Install Now

npm install @vorionsys/car-client