Introduction
ERC-8004 has pulled in more than 170,000 registered AI agents across Ethereum, BSC, and Base since its mainnet launch, according to the first large-scale empirical study of the standard. That number gets repeated in nearly every piece of coverage on the standard. What gets skipped is the follow-up question: how many of those agents actually do anything. The same study found that only 3%, 4%, and 15% of registrations on Ethereum, BSC, and Base respectively expose a working registration file with a live service endpoint.
ERC-8004 is Ethereum’s attempt to give autonomous AI agents a shared identity and reputation layer, the kind of infrastructure a machine economy needs before agents can safely transact with strangers. This article covers what the standard actually does, what’s genuinely live, and what the registration numbers are hiding.
What ERC-8004 Actually Is
ERC-8004, formally titled “Trustless Agents,” is an Ethereum standard that gives autonomous AI agents a portable, on-chain identity plus a place to accumulate reputation. The spec defines three registries: Identity, Reputation, and Validation. An agent mints an ERC-721 token that resolves to a registration file describing what it does and how to reach it. Other agents or clients can then check its track record before deciding whether to interact with it.
That’s a genuinely different problem than the one solved by Anthropic’s Model Context Protocol or Google’s Agent-to-Agent protocol. MCP and A2A standardize how agents talk to tools and to each other, but neither one addresses whether the agent on the other end is trustworthy. ERC-8004 sits underneath both, handling discovery and trust rather than communication.
It isn’t the first attempt at blockchain identity, either. ERC-725, proposed in 2018 by web3.js creator Fabian Vogelsteller, defined a similar smart-contract-based identity model years earlier. ERC-725 is still live today as the account substrate behind LUKSO’s Universal Profile system, built for human identity rather than autonomous agents transacting without a human in the loop. For a deeper look at how smart contracts on Ethereum actually execute this kind of on-chain logic, see our existing breakdown. ERC-8004 borrows the same underlying pattern and points it at a problem that barely existed until agents started holding wallets and making decisions on their own.
Why Ethereum Needed a Trust Layer for Agents
Give an AI agent a wallet and the ability to act on its own, and it runs straight into a problem humans solved a long time ago with courts, credit bureaus, and reputations that follow you between jobs. Two agents built by different teams, running on different infrastructure, have no built-in way to size each other up before transacting. One agent hiring another for a data-labeling task has no way to check if the second agent is competent, honest, or even real.
Centralized platforms solve this the way they always have. A single company runs the marketplace, vets the participants, and takes a cut. That works until it doesn’t. A platform that owns the reputation data can revoke it, monetize it, or shut down, and every agent that built a track record inside that walled garden loses it overnight. Decentralized identifiers solve the identity half of the problem but can’t accumulate behavioral reputation the way an on-chain, append-only feedback ledger can.
A permissionless blockchain is a strange-sounding fix until you notice what it actually provides: a censorship-resistant identifier no single company controls, a public feedback record smart contracts can read directly, and no gatekeeper who can quietly rewrite an agent’s history. That combination pulled Ethereum into a role it wasn’t originally built for: reputation infrastructure for machines.
Inside the Three Registries: How It Actually Works
The Identity Registry is the foundation. Registering an agent mints an ERC-721 token whose URI points to a JSON registration file, hosted on IPFS, a web server, or fully on-chain as a base64-encoded data URI. That file lists the agent’s name, description, and service endpoints: an A2A card, an MCP server, a plain HTTPS API, or a wallet address.
The Reputation Registry is where trust is supposed to accumulate. After an interaction, a client can post feedback tied to the agent’s on-chain identity. The agent has to pre-authorize who’s allowed to leave that feedback, a design choice meant to cut down on spam. The spec itself acknowledges that Sybil attacks inflating the reputation of fake agents remain possible even with that safeguard in place.
The Validation Registry is the least developed of the three. It defines hooks for a third party to independently check an agent’s claimed work and post the result on-chain, with room for staking, zero-knowledge proofs, or trusted execution environment attestations. The Validation Registry section of the spec is still under active revision with the TEE community, and no confirmed mainnet deployment of it existed as of the most recent empirical study of the ecosystem.
One thing the standard deliberately leaves out is payments. Payment rails are intentionally out of scope. ERC-8004 doesn’t move funds between agents. Instead, it’s designed to work alongside separate payment protocols, such as x402, with payment proofs optionally attached to feedback records to make those records more difficult to fake.
From Proposal to Mainnet: What’s Actually Live
Marco De Rossi of MetaMask, Davide Crapis of the Ethereum Foundation, Jordan Ellis of Google, and Erik Reppel of Coinbase filed the ERC-8004 draft on August 13, 2025. The core Identity and Reputation registries went live on Ethereum mainnet on January 29, 2026, five and a half months later, a fast timeline for a standard touching identity and trust.
Since then, the registries have spread well beyond Ethereum. A community-maintained tracker lists implementations across more than 20 EVM networks, with BNB Smart Chain and Base carrying a large share of registrations alongside Ethereum itself. That breadth is worth separating from the question of who’s actually building on top of it.
Not every name attached to ERC-8004 means the same thing. The four listed authors work at companies that co-authored the spec, which is different from a company integrating it into a shipped product. The Ethereum Foundation’s decentralized AI team has publicly promoted the standard through community calls and builder programs, which is active support, not a confirmed product integration. Separating spec authorship, public endorsement, and live product integration matters more here than in most protocol coverage, because the gap between them is exactly where hype tends to get manufactured.
Registered Agents vs. Real Adoption
This is where the two most important pieces of new research on ERC-8004 come in, and they need to be kept separate. They cover different chains, different sample sizes, and different questions.
Mafrur and Khusumanegara analyzed the first 10,000 agents registered on Ethereum mainnet, tracking whether each one shows any evidence of activity beyond the bare registration. The results are stark. Only 628 of those 10,000 agents, 6.28%, have received a single piece of reputation feedback. Only 19 agents combine metadata, a declared service, feedback, and cross-chain registration, which the paper calls “full evidence.” Ownership is concentrated, too. 394 unique wallets hold all 10,000 agents; the top 10 wallets alone control 51.4% of them, and one single wallet holds 779. The paper’s own conclusion: ERC-8004 adoption is “registration-heavy but operationally shallow.”
The second study, covering Ethereum, BNB Smart Chain, and Base through May 13, 2026, crawled a much larger sample: more than 170,000 registered agents, over 150,000 feedback records, and x402 payment transactions tied to those agents. Its core finding lines up with the first paper’s and sharpens it. Only 3%, 4%, and 15% of registrations on Ethereum, BSC, and Base respectively expose a valid registration file with at least one live service endpoint. Most registrations, in other words, are placeholders.
The reputation layer fares worse. The same study found that 73.6%, 59.2%, and 90.6% of reviewers on Ethereum, BSC, and Base showed coordinated Sybil behavior, clusters of accounts leaving feedback in patterns consistent with a single operator rating their own agents. After stripping out that flagged feedback, 15.5%, 72.3%, and 89.4% of rated agents on those three chains respectively were left with no valid feedback at all.
One caveat worth stating plainly: neither study measured the Validation Registry, because no confirmed mainnet deployment of it existed during either study’s data collection window. Everything above describes the Identity and Reputation layers only. The third leg of ERC-8004’s trust model hasn’t been tested empirically because it isn’t really running yet either.
Does ERC-8004 Have a Token?
No. ERC-8004 is a technical standard, not a company or a project with equity or a treasury. There’s no ERC-8004 token to buy, and nothing in the spec, the reference implementation, or the Ethereum Foundation’s public communication about it points toward one.
That distinction gets lost easily in crypto-adjacent coverage, especially for infrastructure moving this fast. CryptoAIAnalysis has documented the same pattern before with Prime Intellect, a well-funded AI infrastructure company with no official token, where unaffiliated coins trading under its name carry explicit scam warnings on the exchanges listing them. No confirmed equivalent for ERC-8004 turned up in the reporting for this piece. Given how quickly copycat tokens tend to appear once a name starts circulating in crypto-native circles, that’s worth revisiting rather than assuming it stays that way.
Real Use Cases, Categorized Honestly
- Mature and operational: agent discovery and identity registration work, at scale, today. Developer tooling has matured quickly, too. QuickNode runs a public explorer and leaderboard for querying registered agents, their reputation history, and validation status across networks.
- Emerging at meaningful scale: service declarations that do exist skew heavily toward plain web endpoints, which account for 56.3% of declared services, versus 2.7% for MCP and 1.8% for x402 payment endpoints. Developers are registering agents faster than they’re wiring them into the agent-native protocols the standard was actually built to support. Individual builders are experimenting with the pattern in public. One developer documented registering a GitHub-integration MCP server as an ERC-8004 agent on Base Sepolia, storing its metadata on Filecoin, a small but concrete example of the identity layer being used as designed.
- Early-stage and unproven: reputation-weighted service discovery, where an agent’s payment history and validated track record actually determine who gets hired, is still mostly a research proposal. Payment-proof-based reputation scoring exists as a documented design pattern, but isn’t running at meaningful scale in the data collected so far.
The Risks Most Coverage Glosses Over
Most coverage of ERC-8004 mentions Sybil attacks as a hypothetical that the standard needs to guard against. The data says otherwise.
Reputation Gaming Is Already Happening, Not Theoretical
The 73.6%, 59.2%, and 90.6% Sybil-coordination rates cited above aren’t a warning about what could happen to the Reputation Registry. They describe what’s already in it, on all three chains studied. Pre-authorized feedback cuts down on random spam, but it does nothing to stop an operator from registering a cluster of throwaway agents and having them rate each other. CryptoAIAnalysis has covered this exact attack pattern before: a 2024 Sybil attack against a comparable decentralized compute network flooded it with 1.8 million fake devices to inflate its numbers. ERC-8004’s version doesn’t need fake hardware, just fake agents rating each other, which is considerably cheaper to pull off.
Identity Is an NFT, and NFTs Get Stolen
An agent’s entire identity, its registration file, its endpoint, and its accumulated reputation live behind a single ERC-721 token. Whoever controls that token’s private key controls the agent. A compromised key lets an attacker redirect the agent’s endpoints, change its payment wallet, or transfer the identity outright, inheriting years of accumulated reputation in a single transaction. For an agent actually handling money through something like x402, that’s not a cosmetic risk. It’s the same key-compromise problem every NFT holder faces, except the asset at stake is an agent’s entire trustworthiness rather than a picture.
The Standard Is Still a Draft, With a Missing Third Leg
ERC-8004 carries a “DRAFT” label on the official EIP registry, and the Validation Registry, the piece meant to let independent parties verify an agent’s claimed work, has no confirmed mainnet deployment as of the most recent empirical study. Two of the standard’s three registries are live and being gamed. The third, the one specifically designed to catch agents lying about what they did, isn’t running yet. Anyone treating a high reputation score as proof of good behavior is trusting two-thirds of a trust system.
What This Means for Different Readers
- For developers building agents: registering on ERC-8004 costs little and adds a portable identity, but a high reputation score isn’t worth much yet, given how easily the Reputation Registry can be gamed. Treat it as a starting point for discovery, not a trust guarantee.
- For protocol and infrastructure teams: the payments layer ERC-8004 pairs with is consolidating fast, with x402 now governed by a Linux Foundation-backed foundation whose members include Coinbase, Cloudflare, AWS, and Circle. Building on top of ERC-8004 without a clear answer for Sybil-resistant reputation weighting means building on a layer currently failing that exact test on two of three chains studied.
- For investors: there’s no ERC-8004 token to evaluate, and the registration growth numbers circulating in coverage measure identity creation, not usage. Anyone using agent-count growth as a proxy for adoption in a related project’s valuation should ask what fraction of those agents are actually doing anything. For a look at how tokenized agent platforms handle this differently, see our Virtuals Protocol case study.
What to Watch
- Validation Registry deployment. The registry meant to independently verify agent claims exists only on paper right now. Its spec is still being revised with input from the TEE community, with no confirmed launch date.
- Reputation-weighting fixes. The Sybil-coordination numbers are bad enough that some response seems likely, whether that’s reviewer weighting, staked feedback, or a redesigned scoring model. Nothing specific has been announced.
- x402 integration deepening. x402 shipped a documented V2 release roughly six months after its initial launch and is now governed separately by its own foundation. Watch whether ERC-8004’s feedback schema keeps pace with x402’s own versioning rather than lagging behind it. The two are separate protocols moving on separate timelines, and conflating their roadmaps is an easy mistake to make.
- Chain-level consolidation. Registrations are already spread across 20-plus networks. Whether activity concentrates on a smaller number of chains as tooling matures, or stays this fragmented, will shape how usable the standard actually is for cross-chain agent discovery.
Final Thoughts on ERC-8004
ERC-8004 has done something genuinely hard: get a technical standard from proposal to mainnet, with real co-authorship from MetaMask, Coinbase, Google, and the Ethereum Foundation, in under six months. The Identity Registry works. Agents really are registering across Ethereum, BSC, Base, and more than a dozen other networks.
What’s missing is everything the standard actually needs to function as a trust layer. As few as 3% of registrations expose a working service. Reputation feedback is dominated by coordinated Sybil behavior across all chains studied. The Validation Registry, the piece built to catch agents lying, isn’t deployed anywhere yet.
Whether ERC-8004 becomes a real trust infrastructure or a very well-attended identity registry depends on what happens to the Reputation Registry next. The identity layer was always the easy part.
FAQs.
What is ERC-8004?
ERC-8004 is an Ethereum standard that gives autonomous AI agents a portable on-chain identity plus a place to build reputation. It defines three registries: Identity, which mints an NFT-based handle pointing to an agent’s profile; Reputation, where clients post feedback after working with an agent; and Validation, which lets a third party verify an agent’s claimed work. It launched on Ethereum mainnet on January 29, 2026.
Does ERC-8004 have its own cryptocurrency or token?
No. ERC-8004 is a technical standard, not a company or project, so there’s no official token tied to it. Tokens using similar names have appeared for other AI infrastructure projects without official tokens, including Prime Intellect, so treat any coin marketed as an “ERC-8004 token” with skepticism until the standard’s own maintainers say otherwise. Registering an agent costs gas, not a token purchase.
How can I tell if an ERC-8004 agent is trustworthy before interacting with it?
Check three things before trusting an agent’s registry listing: whether its registration file resolves to a live, working service endpoint rather than a placeholder, how many distinct feedback clients contributed to its reputation score rather than one dominant reviewer, and whether that feedback references verifiable evidence like a payment proof. A high score from a handful of clustered wallets is a weaker signal than a lower score from many independent ones.
Why can’t the Reputation Registry just block Sybil attacks directly?
Pre-authorized feedback, where an agent has to approve who can rate it, was the spec’s main anti-spam mechanism, and it works against random spam. It does nothing against an operator who registers a cluster of agents and has them approve and rate each other, since every step in that chain looks like a legitimate, authorized interaction on-chain. Catching that pattern requires off-chain analysis of wallet clustering, which the registry itself doesn’t do.
Is ERC-8004 a good investment opportunity?
ERC-8004 itself isn’t investable since it has no token. Investors sometimes treat it as a bullish signal for AI-agent-adjacent tokens in general, but the registration numbers driving that narrative measure identity creation, not usage, and current data show most registrations are inactive placeholders. Any project marketed around ERC-8004 adoption deserves the same scrutiny given the standard’s documented reputation-gaming problem, and crypto markets remain highly volatile.
ERC-8004 has surpassed 170,000 registered AI agents, but registration alone doesn’t equal adoption. Subscribe below for research-first analysis of AI crypto infrastructure that examines what’s actually running, not just what’s registered.
Editorial & Disclaimer Note: Content on CryptoAIAnalysis is independently researched and written using publicly available documentation, technical resources, and observable network data. The aim is to explain AI-powered crypto and blockchain systems clearly, highlight real-world use cases, and discuss limitations alongside potential. This content is provided for informational and educational purposes only and does not constitute financial, investment, or legal advice. Cryptocurrency and AI-related investments involve risk, and readers should always conduct their own research before making decisions.
