We’d love your feedback.
Take a 30-second survey to help improve The Block.
Ethereum announced that ERC-8004, a proposal defining a standard for trustless AI agents on the network, is set to go live on mainnet soon.
The proposal, introduced in August 2025, aims to allow AI agents to interact with different organizations and platforms on Ethereum to enable a decentralized, permissionless economy where agents act as full economic participants.
“By enabling discovery and portable reputation, ERC-8004 allows AI agents to interact across organizations ensuring credibility travels everywhere,” Ethereum wrote on social media platform X on Tuesday. “This unlocks a global market where AI services can interoperate without gatekeepers.”
The proposal said AI agent trust models are pluggable and tiered by risk and security level, allowing agents to perform low-stake tasks like ordering pizza to high-stake cases such as medical diagnosis.
Davide Crapis, the AI lead at the Ethereum Foundation, said the network is in a “unique position” to become the platform that settles AI-to-AI interaction. Part of Ethereum’s mission is to connect the AI community with decentralized infrastructure, Crapis added.
While the official Ethereum account did not provide a specific launch date, Marco De Rossi, head of AI at MetaMask and one of the co-authors of the proposal, said the mainnet launch will take place “probably around Thursday 9 am ET.”
Identity, reputation, validation
ERC-8004 establishes three lightweight smart contract registries, deployable on mainnet or any Layer 2s, as a mechanism to discover and trust AI agents in untrusted settings.
The identity registry provides every agent with a portable, censorship-resistant identifier, which makes every agent browsable and transferable with NFTs-compliant apps. The reputation registry provides an interface for signed feedback from clients, such as agent ratings. The validation registry allows agents to request verification of their work and allows validator smart contracts to provide responses that can be tracked on-chain.
Meanwhile, ERC-8004 also laid out potential risks related to security. The proposal cited the possibility of Sybil attacks, where a malicious actor creates a large number of fake identities to gain disproportionate influence.
It also noted that the ERC cannot cryptographically guarantee that advertised capabilities of an agent are fully functional and non-malicious, but instead relies on three trust models — reputation, validation, and trusted execution environment attestation — to mitigate this risk.
