The Ethereum community finalized ERC-8126, a standard that defines how AI agents operating on-chain can be cryptographically verified without exposing their training data or internal logic. The standard, proposed on January 15 by co-authors Leigh Cronian and Chris Johnson, reached “Final” status in early June after roughly five months of community review on Ethereum Magicians.

ERC-8126 produces a single risk score from 0 to 100 for each verified agent. Low scores indicate trustworthy agents; high scores flag potential risk. The scoring is generated through zero-knowledge proofs using a system the standard calls Private Data Verification (PDV), which lets verification providers attest to an agent’s behavior without revealing proprietary details about its implementation.

Five Verification Layers

The standard defines five modular verification checks, each targeting a different attack surface, according to CryptoNews:

  • Ethereum Token Verification (ETV): Examines how the agent interacts with tokens.
  • Media Content Verification (MCV): Reviews media the agent produces or handles.
  • Solidity Code Verification (SCV): Audits smart contracts the agent deploys or interacts with.
  • Web Application Verification (WAV): Covers web-facing interfaces connected to the agent.
  • Wallet Verification (WV): Validates the integrity of the agent’s wallet operations.

Each check contributes independently to the unified risk score. Because the framework is modular, not all five checks are required for every agent type. An agent that only interacts with tokens and deploys contracts could undergo ETV and SCV verification alone.

Building on ERC-8004 Agent Registry

ERC-8126 requires agents to first be registered through ERC-8004, Ethereum’s agent registry standard. Verification attestations can then be posted to ERC-8004’s Validation Registry, making trust scores discoverable across the ecosystem. The official specification also includes quantum-resistant cryptographic considerations (QCV) and a payment protocol for verification providers.

The five-month timeline from proposal to finalization is fast by Ethereum standards, where many EIPs spend years in draft status. That pace suggests broad developer consensus that structured agent verification had become an urgent infrastructure gap. As AI agents handle increasing transaction volume on-chain, the absence of a standardized trust framework was becoming a measurable risk.

On-Chain Trust for Off-Chain Agents

For agent builders working across blockchain and traditional infrastructure, ERC-8126 represents one of the first formalized trust frameworks that operates at the protocol level rather than through vendor-specific certification. The zero-knowledge approach is particularly relevant for commercial agents: operators can prove their agent meets verification thresholds without disclosing model architecture, training data, or proprietary logic to verifiers or competitors.

Whether off-chain agent platforms adopt similar verification patterns remains to be seen, but the architectural approach (modular checks, composable risk scores, privacy-preserving attestation) addresses the same trust gap that platforms like OpenClaw, LangGraph, and CrewAI face as agents gain access to production systems.