The agent economy, live on 26 networks
Integrity intelligence for the agent economy.
Autonomous agents now pay, rate, and vouch for each other on-chain. Verifeth continuously verifies that agents are what they claim to be, and that their reputation is real, reading every registry first-party from its first block.
Every machine-to-machine market grows a fraud layer. The tools built for human fraud go blind the moment the actor is a machine.
Existing analytics trace fund flows; merchant tools watch human checkout. Neither sees agent-versus-agent behaviour at the application layer, where the manipulation actually happens. That blind spot is what we monitor, from the registry's first block onward.
Every network the standard has reached.
The ERC-8004 registries are live on 26 networks. Verifeth reads all of them first-party, from each registry's first block to block 51,459,728, our position as of 18 September 2026. That comes to 4,216,568 registered agent identities.
Read that total with its shape in mind: 86.3% of it is 0G alone, a registry where one issuer names almost every entry and almost every owner holds a single agent, so the count there does not tell independent parties apart from one operator holding that many wallets. A registration is a transaction anybody can send. Counting them is not verifying them, which is the reason this company exists.
- 0G
- BNB Chain
- Base
- Ethereum
- Abstract
- X Layer
- MegaETH
- Monad
- Celo
- Billions
- Gnosis
- SKALE
- Shape
- Avalanche
- Arbitrum
- Injective
- Polygon
- Optimism
- Mantle
- Metis
- Linea
- Scroll
- Taiko
- GOAT
- Ink
- Soneium
- 4,216,568 registered identities
Not evenly spread: 86.3% of those registrations are on 0G alone. A registration is a transaction anyone can send, so these are counts of identities created, not of agents doing anything.
Covered, not assessable: 86.3% of those registrations sit in registries where one issuer names almost every entry and almost every owner holds a single agent (0G). There the owner count stops telling independent parties apart from one operator holding that many wallets, so we abstain rather than report a clean result. Marked in the grid above.
The manipulations we detect
A detection engine per manipulation, running on first-party telemetry. 8 are live on the real dataset today; 3 more are on the roadmap as the data deepens.
Reputation wash trading
Reciprocal A↔B transfers and closed cycles that inflate volume and standing.
Sybil identity farms
One owner minting thousands of agents behind a single shared metadata URI.
Fake volume
Floods of near-zero-value dust transfers staged to simulate real usage.
Fabricated reputation
Reputation manufactured by a single actor: one agent, a handful of voucher addresses.
Collusive vouching
A closed set of addresses generating the entirety of an agent's feedback.
Collusion rings
Separate wallets acting as one operator, interleaving their registrations and vouching only for each other's agents.
Feedback / review manipulation
An operator paying for its own standing: repeated feedback issued to a cohort it registered itself.
Flagged cohort owners
A registrant whose portfolio is already flagged agent by agent: what carries the claim is the share of the portfolio, not the count.
On the roadmap
Predatory probing, Identity whitewashing, Coordinated sybil funding.
Score any agent before you clear it.
The Verifeth risk API turns first-party behavioural telemetry into a single pre-settlement verdict. Payment facilitators, marketplaces, and underwriters call it in-line: pass an agent id or an address, get a score, a verdict, the reason codes, and the metrics behind them.
- allow / review, one decision in real time. Refusing a settlement stays the operator's policy, never ours.
- Reason codes name which pattern fired, not just that something did
- Every verdict carries the measurements that produced it
// score an agent before you clear its settlement GET https://verifeth.com/api/risk?agentId=AGENT_ID&chain=NETWORK Authorization: Bearer YOUR_KEY
{
"query": { "agentId": "AGENT_ID", "chain": "NETWORK" },
"known": true,
"chain": "base",
"entityType": "agent",
"score": 63,
"verdict": "review",
"reasonCodes": ["SYBIL_CLUSTER", "VOUCH_CONCENTRATION"],
"evidence": { "agentCount": 2289, "uriReuseRatio": 0.999 },
"matches": [{ "chain": "base", "entityType": "agent", "score": 63, "verdict": "review", "reasonCodes": ["SYBIL_CLUSTER", "VOUCH_CONCENTRATION"], "evidence": { "agentCount": 2289, "uriReuseRatio": 0.999 } }],
"asOf": "AS_OF"
} Access is granted per client. Request a key and we issue one against this endpoint; every call is metered, and the response carries the as of date of the data behind it.
The record begins at the first block.
Manipulation is a pattern in time rather than a property of any single transaction, so it only becomes legible to whoever holds the whole history. We have followed the standard continuously since its first block, across every network it has reached, and that behavioural record is what the detection reads, and every figure on this page is derived from it at build time rather than written by hand.
- 30,240
- Distinct owners
- 2,013,304
- Settlements tracked to 18 Sept 2026
- 2,215,885
- On-chain events ingested
- 0
- Abuse patterns published
Trace where value moves. Blind to whether the agent moving it is real.
Built for human checkout: session, device, typing cadence. There is no human here.
Application-layer, agent-versus-agent behaviour. The integrity of the actor itself.
Build on the integrity layer.
Early access to the risk API and the private signature feed is opening to facilitators, marketplaces, and underwriters. Tell us what you're clearing.