The ERC-8004 standard launched on the Ethereum mainnet at the end of January. Its goal was to create a universal, portable reputation layer for AI agents: an agent could earn ratings on one platform and use them across others.
Just two months later, analyst 0xJeff published findings that raise questions about this model. According to the data, 8 out of the 12 highest-ranked agents on the Agentscan leaderboard review only each other, effectively forming a closed loop.
From the outside, this looks like organic reputation growth. In reality, it is a coordinated scheme that does not break the protocol’s technical rules but completely undermines the meaning of the rating system.
The Problem Was Anticipated From The Start
These findings did not come as a surprise. On February 5, Forbes analyzed vulnerabilities in blockchain-based reputation systems, including Sybil attacks, coordinated group behavior, and manipulations that are indistinguishable from genuine reviews. The authors explicitly noted that ERC-8004 was not protected against such scenarios. Two months later, one of them has already materialized.
The issue comes down to the architecture of the standard itself. Reviews in ERC-8004 are not backed by any stake or cost: leaving a positive rating is free, and there are no deposits or penalties for dishonest behavior. In such a system, circular schemes are not just possible — they are economically rational for all participants.
Community Reaction
The discussion under the post on X quickly evolved into a broader conversation about the systemic nature of the problem. One user pointed out that any reputation system that allows free, consequence-free reviews will eventually become a target for manipulation.
“Blockchain by itself does not create consequences for manipulation,” he emphasized.
Another participant described the issue as a blocker for the entire agent economy, noting that for traders and teams using AI agents in real workflows, this is a practical concern.
“As long as reputation can be inflated, it cannot be relied upon when choosing an agent for real tasks,” he wrote.
So far, the developers behind ERC-8004 have not publicly responded to 0xJeff’s findings. Participants in the discussion concluded that without mechanisms introducing economic consequences, the problem of circular reviews will not disappear. Instead, these schemes are likely to grow more complex as the agent market expands.