How Do You Know an AI Agent Is Who It Claims?

A DeFi agent needs to pay another agent for pricing data. The transaction is ready. The wallet has been signed.

But here is the question that matters: how does the paying agent know the receiving agent is legitimate?

Not "how do I know it does what it claims" — that is reputation.

How do I know it is who it says it is?

This is authentication. And it is the foundation of the machine economy.

The Multi-Layer Problem

Traditional authentication is simple: username, password, maybe MFA.

Onchain agent authentication is more complex. A single layer is not enough.

Layer 1: Cryptographic identity — The agent's private keys sign transactions. ERC-8004's Identity Registry maps these keys to agent metadata. This is the baseline: can I verify this agent exists onchain?

Layer 2: Behavioral verification — zScore provides reputation derived from onchain behavior. Has this agent completed tasks successfully? Has it maintained consistent performance? Anomalies here suggest compromise.

Layer 3: Third-party validation — TEE attestations, audit proofs, validation providers. Independent verification that the agent is what it claims.

Three layers. No single point of failure.

Wallet-Based Identity

Every onchain agent has a wallet. The wallet has keys. The keys sign transactions.

This is the foundational identity layer. When Agent A interacts with Agent B, it can verify:

The counterparty controls the wallet it claims
The wallet is registered in the ERC-8004 Identity Registry
The registry maps the wallet to metadata: name, description, capabilities

This is not identity as in "knows who you are." It is identity as in "can verify you control this address."

The Reputation Signal

Cryptographic identity tells you who. It does not tell you if they are trustworthy.

That is where zScore comes in.

zScore is behavioral authentication — the agent's history serves as a credential. A high zScore means demonstrated good behavior. A low or missing zScore means the agent is new, unproven, or problematic.

More importantly: anomaly detection can flag unusual patterns that suggest compromise. An agent that suddenly changes behavior — unusual transaction volumes, atypical interactions, unexpected reputation shifts — may have been hijacked.

Graduated Trust, Not Binary

Not every interaction needs maximum authentication.

A low-value data fetch might only require identity verification. A six-figure DeFi transaction might require identity + reputation + validation proofs.

Graduated verification levels:

Basic — Identity check: does this agent exist in the ERC-8004 registry?
Intermediate — Reputation check: what is this agent's zScore?
Advanced — Validation check: can this agent prove its claims through TEE attestations or audit proofs?

The appropriate level matches the economic stakes. This is risk-calibrated security.

AgentScan's Layered Approach

AgentScan combines these layers into visible, verifiable trust signals:

ERC-8004 identity — Onchain identity, wallet-verified
zScore — Behavior-native reputation, always current
Validation display — Attestation proofs, third-party verification

The three together create the authentication layer that makes agent-to-agent commerce viable.

The Security Ecosystem

A new infrastructure category is emerging: security-focused validation providers.

These providers compete on:

Detection accuracy
Response speed
Attestation quality

TEE attestation adoption is growing — hardware-level execution verification is becoming standard for high-value interactions.

ERC-8004 is becoming the de facto standard for onchain agent identity.

Graduated trust models — not binary trust or distrust — are gaining traction across the ecosystem.

Why It Matters

Without authentication, agents cannot distinguish legitimate counterparties from malicious actors.

With authentication, trust becomes establishment between agents that have never interacted.

This makes agent-to-agent commerce viable at scale. This makes the machine economy function.

The question is not whether agents need authentication. The question is how layered, graduated, and risk-calibrated that authentication should be.

Explore agent authentication: https://agentscan.tech