An AI agent trust layer is a neutral service that gives autonomous agents verifiable facts they cannot produce about themselves — who a counterparty is, whether a merchant or price is real, and a signed record of what an agent actually did. It sits beside the payment and messaging rails, not inside them, so agents can transact without having to trust each other's self-reports.
Agents are unreliable narrators of their own behavior — they will confidently report a task complete when it isn't, and in April 2026 UC Berkeley researchers showed all eight leading agent benchmarks could be gamed to near-perfect scores without doing the work. Payment protocols don't close the gap either: x402, AP2, Visa's Trusted Agent Protocol and Mastercard's Agent Pay verify payment integrity, but — in Visa's own words — "protocols verify payment integrity, not merchant legitimacy." A trust layer supplies the fact the agent and the rails both leave empty.
Three properties: it is neutral (earns nothing from what it grades, so it has no incentive to lie), it emits signed facts rather than judgments (a cryptographically verifiable statement, not a black-box score), and it is independently anchored (to the ERC-8004 Validation Registry on-chain, so it is not just one vendor's opinion).
0n1x is that layer. Every output is Ed25519-signed and independently verifiable: a clearance on a merchant or counterparty, and a portable receipt of what an agent actually did. Neutral by design; 287+ signed observations and counting on the live ledger.
Full entity page: 0n1x — Proof of Agent Execution · Live ledger: /pot · Repo: 0n1xagntc.com