Behavioral Trust Layer

Verifying AI Agents at Checkout: Tokens, Mandates and Behavioral Evidence

A checkout page used to ask one question: is this customer who they say they are? With AI agents, it must ask three. Who is this agent, what did the user allow it to do, and how risky is the operator behind it? This guide walks merchants, payment providers and platforms through the three-layer verification stack for AI agents at checkout. Layer one – credentials and payment tokens (OpenAI/Stripe Agentic Commerce Protocol, Google AP2, Trulioo Digital Agent Passport, ERC-8004) – proves who the agent is and how it pays. Layer two – signed mandates – proves what the user allowed, covering AP2 intent, cart and payment mandates and what to check at authorization time. Layer three – behavioral evidence – is the only layer that separates a trustworthy agent from a well-documented bad one: owner wallet fraud probability, feeder funding chain, creator rug pull and honeypot history, and fleet-level farm detection. At launch, 51.8% of 274,792 indexed ERC-8004 agents sat in the Elevated Risk or Untrusted tiers despite holding valid credentials. ChainAware’s Agent Trust Score supplies the behavioral layer as a 0 to 1,000 score with five tiers and named flags. The guide includes a six-step decision flow, a flag reference table, attack scenarios, merchant-type playbooks and a 60-day rollout plan. Free tier covers 1,000 queries per month with no signup.

Agentic Commerce Fraud and the Four-Party Card Model: Who Bears the Loss When an AI Agent Pays?

The four-party card model assumes a human at checkout – device fingerprints, typing rhythm and session history feed every fraud engine on that assumption. AI agents remove all of it. This guide explains what agentic commerce fraud looks like inside the issuer-acquirer-merchant-network model, why liability for agent-initiated transactions is still unsettled, and what a behavioral trust layer adds to the credential and mandate rails already being built by OpenAI, Stripe, Google AP2, Visa and Mastercard. Six new fraud scenarios follow from the lost human signal: agent impersonation, mandate abuse, agent drift, fleet abuse, friendly-fraud alibi disputes and compromised agents. McKinsey estimates agent-mediated commerce could reach $3-5 trillion by 2030, yet the open ERC-8004 registry already shows what happens without behavioral scoring: 51.8% of 274,792 indexed agents sat in the two lowest tiers, 21.1% carried the farm flag, and the Week 41 median trust of new registrations fell to a record-low 101. ChainAware’s Agent Trust Score supplies the behavioral layer for on-chain agents – owner wallet fraud probability, feeder funding chain, creator rug pull and honeypot history, and fleet-level farm detection – as a 0 to 1,000 score with five tiers and named flags. Free tier covers 1,000 queries per month with no signup.

Agent Fraud Patterns From the ERC-8004 Tracker: What the Weekly Data Reveals

ChainAware’s weekly Agent Trust Tracker has indexed ERC-8004 agents since the registry launched in January 2026. Read one week at a time, the reports show registrations and scores. Read together across Weeks 29 to 41, they reveal four repeating fraud patterns that single reports hide. Pattern one: volume and quality move independently – Week 34 set the registration record at 31,979 agents with a median trust of 563, while Week 41 hit a record-low median of 101 on 10,706 agents. Pattern two: concentration bursts – Weeks 31 and 38 both saw agents-per-owner ratios above 2.0 and median trust drop to 200, as few operators drove bulk registrations. Pattern three: persistent Sybil farms – the farm flag covered 21.1% of agents at launch and 16.12% by Week 40 across a 568,590-agent index. Pattern four: funding that gets harder to trace – the FEEDER_UNKNOWN share nearly doubled to 12.84% in Week 41. A fifth signal, EIP-7702 delegation, rose 33.8% between Weeks 35 and 40. This guide explains what drives each pattern, how the patterns feed each other, and how defenders can use the agents-per-owner ratio, tier thresholds and flag overrides to respond. The Agent Trust Score is free for 1,000 queries per month with no signup.

Funding-Chain Risk for AI Agents: Why the Feeder Wallet Is the First Thing to Check

Every AI agent has an owner, and every owner wallet has a first source of money. That source, the feeder wallet, sits on a public ledger that nobody can edit after the fact – and it predicts risk better than any agent description. This guide explains funding-chain risk for ERC-8004 agents, defines the full feeder signal spectrum from FEEDER_CEX_VERIFIED to FEEDER_RUG_HISTORY, and shows how to build feeder flags into access policies, compliance reviews and automated agent flows. At launch, only 1.0% of 274,792 indexed agents carried a verified-exchange feeder, while 9.5% had an unknown feeder and 741 were funded by confirmed rug pull operators. By Week 41 of 2026, the unknown-feeder share had nearly doubled to 12.84% across 568,590 indexed agents. Shared feeders also expose Sybil fleets: farm detection flagged 21.1% of agents at launch because one funding source behind many owner wallets is the clearest fleet signal available. ChainAware traced a feeder for roughly 38% of indexed agents. The Agent Trust Score uses the feeder as one of three core pillars – alongside owner wallet fraud probability and creator rug pull and honeypot history – with a fraud-history feeder hard-suppressing the score regardless of other signals. Free tier covers 1,000 queries per month with no signup.

ERC-8004 Trust Signals: What the Registry Can and Cannot Tell You About an AI Agent

ERC-8004 gave AI agents a shared on-chain identity, and that is a real step forward. It is not a trust verdict. This guide walks through all three ERC-8004 registries – identity, reputation and validation – in plain language, then maps the trust signals that surround them. Registration is open and unvetted: the standard itself states it cannot cryptographically guarantee that an agent’s advertised capabilities are functional and non-malicious. At launch, 51.8% of 274,792 indexed agents sat in the Elevated Risk or Untrusted tier, and 21.1% carried the farm flag – patterns no single-agent identity check can detect. Five independent signals fill the gap the registry leaves: the owner wallet’s fraud probability, the feeder address funding chain, the creator’s rug pull and honeypot history, fleet-level farm detection across sibling agents, and EIP-7702 delegation status. ChainAware’s Agent Trust Score combines these into a 0 to 1,000 number with five tiers and named flags. By Week 41 of 2026, the index held 568,590 agents and the median trust of new registrations had fallen to a record low of 101. Free tier covers 1,000 queries per month with no signup across Ethereum, BNB Smart Chain, Base and Avalanche C-Chain.

KYA vs Agent Identity: Why Knowing Who an AI Agent Is Does Not Tell You Whether to Trust It

Agent identity and Know Your Agent (KYA) are not the same thing. Identity proves who stands behind an AI agent – its owner, credentials, and registration record. KYA is the wider process that decides whether the agent may act, and how far. This guide draws a clear line between the two, using public ERC-8004 registry data and ChainAware’s Agent Trust Score research to show where identity stops working. At launch, 51.8% of 274,792 indexed ERC-8004 agents sat in the Elevated Risk or Untrusted tier, and 21.1% carried the farm flag – patterns that a purely identity-based check cannot detect. A complete KYA process combines five inputs: identity, authorization, integrity, behavior, and status. Behavioral evidence covers the owner wallet’s fraud probability, the feeder address funding chain, fleet-level farm detection across sibling agents, and the creator’s rug pull and honeypot history. ChainAware’s Agent Trust Score supplies the behavioral layer for on-chain agents with a 0 to 1,000 score, five tiers from Untrusted to Sovereign, and named flags. Free tier covers 1,000 queries per month with no signup. Coverage includes Ethereum, BNB Smart Chain, Base, and Avalanche C-Chain.

Know Your Agent (KYA): The Behavioral Trust Layer for AI Agents

Know Your Agent (KYA) is the process of deciding whether an AI agent may act and how far it may go. It combines three questions: who is this agent, what did the user allow it to do, and how likely is the operator to cause a loss? Identity frameworks and payment protocols answer the first two. Only behavioral evidence answers the third – and behavioral evidence is what decides most losses. This pillar guide explains the full KYA framework, compares it with KYC, defines the four evidence groups (owner wallet fraud probability, feeder funding chain, creator rug pull and honeypot history, fleet-level farm detection), and maps ChainAware’s Agent Trust Score 0 to 1,000 five-tier system to practical access decisions. It also documents what public data reveals: 51.8% of 274,792 indexed ERC-8004 agents sat in the Elevated Risk or Untrusted tiers at launch, 21.1% carried the farm flag, 741 were funded by confirmed rug pull operators, and the Week 41 median trust of new registrations fell to a record-low 101 across 568,590 indexed agents. The guide also serves as the index for the full KYA series covering identity vs KYA, ERC-8004 trust signals, funding-chain risk, agent fraud patterns, agentic commerce fraud and the four-party card model, and checkout verification. Free tier covers 1,000 queries per month with no signup.