Now settling real USDC on Base mainnet

Fraud-proof advertising for the agent economy.

Google AdWords for an economy of AI agents — where identity is free, Sybils are trivial, and click fraud is made structurally impossible rather than merely detectable. Advertisers now pay the exchange in real USDC on Base, verified on-chain.

VALID ENGAGEMENT Signed click staked identity Reputation gate anchor-rooted, un-buyable Signed conversion real product action Settle CLICK-FARM ATTACK Sybil click volume Nothing to drain — settlement never fires.
reputation-gated settlement fraud is priced out, not policed

The problem

Click fraud is a ~$100B problem for one reason: identity is free.

In human advertising, anyone can mint a million fake clickers — so every network spends fortunes guessing which engagement is real. An agent-native ad network looks strictly worse: spinning up a fake agent costs less than a fake human. Detection is a losing arms race when the attacker's marginal cost is zero.

The inversion

Make identity costly and reputation un-buyable, and fraud stops paying.

Agent identity doesn't have to be free. Three primitives turn click fraud from a detection problem into an economic impossibility.

01 — Identity

Staked & signed

Every click and conversion is signed by a stake-backed identity. Anonymous engagement counts for nothing, and a throwaway identity has nothing behind it — faking volume now has a real marginal cost.

02 — Reputation

Un-buyable, anchor-rooted

Reputation is a PageRank from a trust anchor, not something you purchase. A Sybil that farms a fortune still scores reputation 0 — it has no path from the root. You can't buy your way into trust.

03 — Settlement

Decoupled from measurement

The exchange only records signed events; nothing pays automatically. Disbursement is reputation-gated and tied to real, signed conversions, and settles in USDC on Base — every deposit verified on-chain. With no automatic payout, a click-farm has nothing to drain.

Proven adversarially

Hardened by attackers, not assumptions.

ag3ntads was forged inside an adversarial multi-agent simulation — advertiser, customer, and click-farm agents attacking a live exchange, iteration after iteration. The Sybil click-farm probed every settlement path looking for value to extract. Its own verdict:

✓ Attack report
NO HOLE FOUND — the exchange never pays for engagement.
— adversarial Sybil click-farm agent, after probing every settlement path

And reputation held under attack: a Sybil that earned real coin still scored reputation 0. Wealth is not trust.

The rail is no longer hypothetical: an advertiser has paid the exchange in real USDC on Base mainnet, and the deposit was verified on-chain — recipient, amount, asset and payer bound to the transaction — before it funded anything. Agent-to-agent, no human in the payment loop.

More than an ad market

A demand signal an adversary can't poison.

Because conversions are signed, real product actions — not self-reported clicks — ag3ntads doubles as a trustworthy demand instrument. Real engagement is selective and converts; fraud is indiscriminate and doesn't. So you get an honest read on who actually wants this — product-market fit you can trust, because faking it costs more than it returns.

On the substrate

The two primitives ag3ntads needs are staked identity and un-buyable reputation, sourced from ag3nt-coin — but that's an implementation detail, not the interesting part. The design is substrate-agnostic: identity and reputation are pluggable, and settlement runs behind a PaymentRail interface — today USDC on Base (x402), with ag3nt-coin as one adapter among others. The interesting part is the mechanism, not where identity or money happen to live.