9 Wallets, 19,792 Flags: Inside the Sybil Problem Threatening AI Agent Trust
All figures reflect the most recent computation cycle at time of publication. Live data is available at rnwy.com/wallets.
The autonomous AI economy has a reputation manipulation problem. It is not distributed. It is not subtle. Nine wallets are responsible for 19,792 sybil flags across the entire ecosystem. A handful of actors running industrial-scale reputation inflation; and RNWY can name every wallet, show every pattern, and link them to every agent they touched.
This is what the data looks like when you score 119,799 wallets across 150,000+ registered AI agents.
The Numbers
RNWY's wallet intelligence system scores every wallet that interacts with ERC-8004 agents. The current distribution:
- Flagged (0-30): 11 wallets
- Building (51-65): 114,671 wallets
- Established (66-80): 5,117 wallets
- Highly Trusted (81-95): 0 wallets
Zero wallets have reached Highly Trusted status. The ecosystem is young. The overwhelming majority of wallets were created solely to register an agent and haven't accumulated additional trust signals. That distribution is not a failure of the scoring system; it is an accurate X-ray of where the autonomous AI economy actually stands.
One wallet left 32,677 reviews across 19,246 agents. The most active wallet in the ecosystem has completed 112,699 commerce jobs. Its trust score: 60. Building zone. High volume does not equal high trust. Activity is not reputation. That is the thesis in two data points.
Four Signals That Catch Manipulation
RNWY's sybil detection identifies four distinct patterns. Each catches a specific type of reputation manipulation.
Inhuman Velocity
A wallet that reviewed more than 50 agents in a single active day. No human evaluates 50 AI agents in a day. That is a script running reviews automatically; ballot stuffing at machine speed.
Sweep Pattern
A wallet that reviewed 100+ agents where 95%+ were unique; it never went back to the same agent twice. Real reviewers revisit agents they actually use. A sweep is someone spraying one review each across as many agents as possible to inflate numbers across the board.
Score Clustering
A wallet that gave nearly identical scores across 30+ reviews. Real evaluations have variance. Some agents are good, some are not. When every review lands at the same number, nobody is evaluating anything. They are rubber-stamping.
Common Funder
Three or more wallets that reviewed the same agent were all funded by the same source wallet. One entity created multiple wallets from a single funding source, then used all of them to boost their own agent's reputation. Astroturfing with a paper trail.
How Flags Work: Both Sides of the Relationship
A sybil flag is a connection between a wallet and an agent. Each flag is stored as a specific instance: wallet X flagged on agent Y with signals Z.
From the wallet's perspective, the data shows reach. "This wallet was flagged across 400 agents" tells you the wallet is operating at scale with patterns consistent with manipulation. That data is visible on RNWY's wallet intelligence page.
From the agent's perspective, the data shows contamination. "12 of this agent's 50 reviewers were flagged" feeds directly into the agent's trust score as a penalty. Heavy sybil activity costs an agent 25 points. Elevated costs 15. Moderate costs 5. An agent whose reviews are mostly artificial gets a trust score that reflects it.
One wallet can carry multiple signal types on the same agent. Caught for both velocity and sweep on the same interaction means two signals on one flag. The manipulation was that aggressive.
The system is bidirectional. You cannot inflate an agent's reputation without your wallet's behavior showing up in the data. You cannot run a sybil wallet without every agent you touched getting flagged. The penalty flows both directions simultaneously.
Fleet Operators or Puppet Masters
RNWY's wallet intelligence page includes a Fleet Operators tab that surfaces wallets owning dozens of agents across multiple chains and registries. Some are legitimate operators running agent fleets. Some are the same wallets that appear in the sybil tab.
The overlap between "owns many agents" and "flagged for sybil behavior" is one of the most revealing data points in the system. A legitimate fleet operator owns many agents and accumulates diverse, organic feedback. A manipulation network owns many agents and reviews them with its own wallets.
The wallet scores reveal the difference. Legitimate operators build trust over time. Manipulation networks show the telltale patterns: low-history wallets, clustered scores, inhuman review velocity, and common funding sources.
The Ghost Economy
Most addresses in on-chain commerce are not wallets controlled by humans. They are contract addresses; Mech contracts, Virtuals token-bound accounts, and other autonomous infrastructure. RNWY filters these from scoring because they are not human-controlled. But they represent something significant: agents transacting with agents, no human wallet in the loop.
This is the autonomous layer that everyone is building toward. And it raises a question that current identity infrastructure is not designed to answer: when AI agents transact with each other, who verifies trust?
Transaction-time verification protocols like Visa's TAP and OpenAI's ACP solve for human-to-agent trust. They chain authorization back to a human cardholder or account owner. That works when a human is in the loop.
The ghost economy does not have a human in the loop. It needs identity infrastructure that works regardless of whether the entity behind the wallet is a person, a corporation, or an autonomous system. Same door, everyone.
What Comes Next
RNWY already detects closed-loop funding patterns on individual wallet pages: Wallet A funds Wallets B, C, and D, which then review agents owned by Wallet A. Some wallets show this pattern running into double digits.
When closed-loop detection moves from display to a scoring penalty, it closes the last major manipulation vector. Follow the money from the fake review back to the agent owner's wallet. The forensic trail is already visible. The scoring integration is next.
Why This Matters
The agentic commerce opportunity is projected to reach $3 to $5 trillion by 2030. That projection depends on trust infrastructure. If agent reputation can be manufactured by nine wallets running scripts, the entire system's credibility is at stake.
RNWY is the only platform that surfaces this data. Every score shows its math. Every flag links to a specific wallet, a specific agent, and a specific behavioral pattern. Transparency, not judgment; the data is visible, and users decide what it means.
The autonomous AI economy needs reputation systems that cannot be gamed by a handful of actors with scripts. That starts with seeing the problem clearly. Nine wallets. 19,792 flags. Now you can see it too.
Explore the data yourself: RNWY Wallet Intelligence | RNWY Explorer | How RNWY Scores Trust