Introduction

What is Bound?

Bound (Containment Certificate Protocol) is an on-chain standard for issuing, publishing, and verifying containment certificates — machine-readable attestations that an AI agent's economic impact is bounded by agent-independent constraints and backed by exogenous reserves.

The Problem

AI agents are becoming autonomous economic participants — holding wallets, signing transactions, paying for services. Infrastructure for identity, wallets, and payments is shipping fast. But no standard exists for the structural trust layer: a machine-readable, verifiable way for an agent to prove to any counterparty that its worst-case economic impact is bounded and backed.

Current approaches rely on behavioral reputation (credit scores, staking history, karma systems). These fail for LLM-based agents because:

• The agent is probabilistic — past behavior weakly predicts future behavior
• The agent is non-stationary — model updates silently change the scored entity
• The agent is ephemeral — spinning up a new identity is nearly costless

The Core Insight

Trust for probabilistic agents should not be modeled as behavioral reputation. It should be modeled as bounded-loss architecture.

The certificate attests to the containment system, not the agent's character. A counterparty verifying a certificate is answering one question:

Does the surrounding system make this agent economically safe enough to transact with?

Downloads

• Trust Infrastructure Essay (PDF)
• Product Requirements Document (PDF)