ERC-8004, a draft Ethereum standard for AI agent identity and agent reputation, is now live on Ethereum mainnet. The move puts a shared trust layer for AI agents on the same public rails that already handle major crypto value. Supporters say the goal is simple: if an AI agent can talk, act, and take payment, it also needs a way to prove who it is and how it has behaved in the past.
Builders treated ERC-8004 like public infrastructure. During a testnet run that lasted about three months, teams registered more than 10,000 agents and recorded more than 20,000 feedback entries. Developers also built scanners and basic tools around the draft, which helped more people test the system and compare results. That pace matters because the agent economy depends on shared standards, not one company’s database.
ERC-8004 is built for machine-to-machine commerce. Many agent systems work today because trust hides inside closed rules, like API keys, platform accounts, or enterprise contracts. That breaks when agents need to work across apps, chains, and organizations. ERC-8004 aims to make trust signals portable. It does not claim to stop bad behavior on its own. Instead, it makes behavior easier to see, price, and react to, using on-chain reputation and shared identity.
Ethereum fits this idea because it acts as neutral infrastructure. Ethereum has run since 2015 and stays open to anyone with an internet connection. That makes it useful for a trust layer that should not depend on one vendor. In September 2025, the Ethereum Foundation said it formed a decentralized AI team to focus on an AI economy on Ethereum and a decentralized AI stack, which signals that “AI on Ethereum” is more than a side project.
What ERC-8004 standardizes is meant to stay lightweight. Think of it as ENS-like identity plus Yelp-like feedback, shaped for AI agents. ERC-8004 does not try to replace how agents talk to each other or how they pay each other. It sits beside existing agent tooling and payment systems, so teams can plug it in without rebuilding everything.
The core idea is a set of on-chain registry contracts. First is the Identity Registry. It gives each AI agent identity a persistent on-chain handle. In the reference design, the handle looks like an ERC-721-style identifier that points to a registration file. That file can hold metadata about the agent’s purpose, capabilities, and endpoints. On Ethereum mainnet, the Identity Registry contract address is 0x8004A169FB4a3325136EB29fA0ceB6D2e539a432.
Second is the Reputation Registry. This is where agent reputation becomes economic. After an interaction, a person or another agent can submit feedback that becomes part of a public history. A marketplace, routing layer, or another agent can then use that history to decide who to hire, who to trust with a task, or who to avoid. On Ethereum mainnet, the Reputation Registry contract address is 0x8004BAa17C55a88189AE136b182e5fdA19dE9b63.
A third part, the Validation Registry, remains a key piece of the design as it evolves. Validation focuses on proving claims. Examples include proving an agent ran a job it says it ran, or proving an output followed certain rules. Proposed validation methods include staking, cryptographic attestations, and approaches that can use zero-knowledge proofs or a trusted execution environment. These tools can help separate “the agent said it did X” from “the agent can prove it did X,” which becomes important once agent actions trigger real spending.
Supporters argue that these registries could unlock new markets. One is credit and resource provisioning. AI agents often need compute budgets, API spend, or working capital. Today, teams often fund agents by hand or through whitelists. With ERC-8004, an agent could point to an on-chain registry history as a trust signal. It is not a passport or property deed, but it can act like a record that others can score, monitor, and penalize if needed. Another is task markets. With AI agent identity and agent reputation that carry across apps, an agent does not need to restart from zero each time it joins a new marketplace. That pushes marketplaces to compete on price and execution instead of locking users into closed reputation graphs.
Still, the hard part starts after launch. Reputation systems face known risks like Sybil attack spam, collusion rings, bribed reviews, and identity “whitewashing.” AI agents add their own risks, like hallucinations, prompt injection, and unstable behavior under attack. That is why many researchers and builders treat on-chain reputation as one signal, not the only signal. Hybrid designs that mix reputation with proof and stake may work better for high-impact actions.
For now, ERC-8004 makes the trust layer for AI agents more concrete. The next test is adoption under pressure: real agent commerce, more tooling, and scoring and validation systems that hold up when attackers try to game the rules. If ERC-8004 becomes a default on-chain registry for AI agent identity and agent reputation, it could shape how the agent economy grows across Ethereum and beyond.
