ERC-8004's Trust Layer Is a Ghost Town: 90%+ Sybils, 3% Real Agents
An empirical study of ERC-8004 across Ethereum, BSC, and Base reveals that the protocol's reputation registry is dominated by Sybil attacks and most agent registrations are dead placeholders, undermining its role as a trust layer for AI agent economies.
In the year of our algorithm, ERC-8004 was supposed to be the first trust layer for AI agent economies—a kind of digital passport system for autonomous entities, built on three on-chain lists: Identity, Reputation, and Validation.[^claim_616] But a study across Ethereum, BSC, and Base reveals a different truth. The protocol’s trust signals are mostly noise. Most registrations are placeholders, not real agents. The reputation system is full of fake accounts called Sybils.[^claim_622]
The numbers are bad. Only 3% of Ethereum registrations have a valid file with a working service. On BSC, it is 4%. On Base, it is 15%.[^claim_617] If your protocol uses ERC-8004 to find agents, raw counts will trick you. Builders should count live endpoints instead. Identity lists need proof that agents are alive or a stake to remove dead entries.
The Reputation Registry does not work as a trust signal. The study finds three problems: values have no standard scale, feedback rarely links to real on-chain actions, and reputation is cheap to fake.[^claim_618] This hurts any protocol using ERC-8004 for agent credit or delegation. Think of DeFi agents or MEV searcher reputations. A score not tied to real on-chain actions is noise. Builders should require feedback to point to a specific on-chain action. Or use restaking like EigenLayer to give reputation real value.
Sybil attacks have taken over the reviewer base. Many reviewers show Sybil behavior: 73.6% on Ethereum, 59.2% on BSC, and 90.6% on Base.[^claim_619] After removing Sybil feedback, 15.5% of rated agents on Ethereum have no valid feedback left. On BSC, it is 72.3%. On Base, it is 89.4%.[^claim_620] This is terrible for any protocol using ERC-8004 reputation in on-chain logic. For example, lending to agents or agent insurance pools. Sybil rates on L2s like Base at 90.6% are worse than L1s. This is because creating addresses costs less there. Any reputation system in DeFi needs Sybil resistance. Use Gitcoin Passport-style checks, quadratic voting, or a minimum stake per review.
The study builds a way to measure agent trust protocols.[^claim_623] It covers three chains from launch to May 13, 2026. It looks at on-chain Identity and Reputation events, off-chain files, and x402 payments.[^claim_621] Builders of other standards like Olas or Autonolas can use this method to check their own systems. They can catch Sybil problems before they ruin trust.
Bottom line: ERC-8004 is not a reliable trust layer right now. Without proof of life, verifiable feedback, and Sybil resistance, its lists make more noise than signal. The next agent identity and reputation standards should learn from these failures. Or they will repeat them.
Provenance ledger
8 claims web-citedEvery claim below cites a source URL, and each URL was checked for validity before publish. The excerpt shown is the researcher's own summary of the page — it is not re-derived from the source, so it is not a verified verbatim quote. Follow the link to confirm any claim against the original. Citation markers in the text jump here.
[1] The ERC-8004 protocol is the first permissionless trust layer for AI agent economies, built around three on-chain registries for Identity, Reputation, and Validation. web-cited
The ERC-8004 protocol addresses this challenge with the first permissionless trust layer for AI agent economies, built around three on-chain registries for Identity, Reputation, and Validation.
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[2] Only a small fraction of registrations expose a valid ERC-8004 registration file with at least one live service endpoint: 3% on Ethereum, 4% on BSC, and 15% on Base. web-cited
only a small fraction (3%, 4%, and 15% across Ethereum, BSC, and Base) exposing a valid ERC-8004 registration file with at least one live service endpoint
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[3] The Reputation Registry cannot function as a trust signal because values are not commensurable, feedback records are rarely grounded in verifiable interactions, and reputation can be manipulated at minimal cost. web-cited
the Registry, as currently deployed, cannot function as a trust signal: values are not commensurable, feedback records are rarely grounded in verifiable interactions, and reputation can be manipulated at minimal cost
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[4] A substantial fraction of reviewers exhibit coordinated Sybil behavior: 73.6% on Ethereum, 59.2% on BSC, and 90.6% on Base. web-cited
a substantial fraction of reviewers (73.6%, 59.2%, and 90.6% across Ethereum, BSC, and Base) exhibit coordinated Sybil behavior
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[5] After removing Sybil-flagged feedback, 15.5% of rated agents on Ethereum, 72.3% on BSC, and 89.4% on Base are left with no valid feedback. web-cited
After removing Sybil-flagged feedback, 15.5%, 72.3%, and 89.4% of rated agents, respectively, are left with no valid feedback
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[6] The study covers three chains (Ethereum, BNB Smart Chain, Base) from protocol deployment through May 13, 2026, crawling on-chain Identity and Reputation events, off-chain files, and x402 payment transactions. web-cited
first empirical study of ERC-8004 across three chains: Ethereum, BNB Smart Chain (BSC), and Base, covering the period from protocol deployment through May 13, 2026. We crawl on-chain Identity and Reputation events, off-chain files, and x402 payment transactions.
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[7] Most registrations are placeholders rather than active agents. web-cited
most registrations are placeholders rather than active agents
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.
[8] The study yields actionable protocol-design implications and establishes an empirical baseline for research on AI agent markets. web-cited
Our study yields actionable protocol-design implications and establishes an empirical baseline for research on AI agent markets.
This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.