| Figure | Value | Reading |
|---|---|---|
| Registrations on Base, end of period | 96,480 | +16,013 minted in the period. The daily cross-check against The Graph's subgraph began after this period; the count was checked against a public Base node instead. |
| Agents seen paying (L7+), end of period | 5,505 | 31 at the start. A wallet bound to the agent paid someone after it registered; not a judgement of anyone. |
| Counted payments in the period | $7,654,905 | 93,673 payments from 7,060 agents to 3,690 addresses; routing hops and pre-registration transfers excluded. |
| Left the payer's control | $4,653,465 | Counted payments minus deposits into positions the payer still holds: the closest figure to spending. |
| First-time payers | 867 | Agents whose first counted payment ever fell in the period. |
| Read late (our own lag) | 73% of value | 86,217 payments ($5,579,419) made in the period were first read by us more than three days later; the per-day figures move as late reads arrive. |
| Poisoning, wallets targeted | 827 | 22,121 look-alike plants; 2 losses ($60). |
| x402 settlements per day, end of period | n/a | Stablecoin transfers into Bazaar-listed sellers in 24 hours; third-party submitted n/a. |
The registry
16,013 registrations were minted on Base in September 2026, taking the Identity Registry from 84,954 to 96,480. Our count is checked against The Graph's Agent0 subgraph at the exact end of every UTC day: 0 of 0 days agreed exactly. The ten largest minting accounts held 29.9% of all registrations at the start and 35.1% at the end; n/a of the agent files in our probe sample sit on one host.
| Figure | Start of period | End | Change | Definition · measured |
|---|---|---|---|---|
| Base registrations (our count from the Identity Registry) | 84,954 | 96,480 | +11,526 | reg.minted.base.v1 · 2026-09-30 21:46 UTC |
| Registrations with a known owner (census) | 95,870 | 96,360 | +490 | census.registrations.owned.v1 · 2026-09-30 02:53 UTC |
| Wallets bound to a registration | 48,816 | 48,995 | +179 | census.bound.wallets.v1 · 2026-09-30 02:53 UTC |
| Distinct minting accounts | 29,812 | 31,188 | +1,376 | reg.minters.v1 · 2026-09-30 23:51 UTC |
| Share of registrations by the ten largest minters | 29.9% | 35.1% | +5.2 pts | reg.top10minter.share.v1 · 2026-09-30 23:51 UTC |
| Live endpoint rate (probe sample) | 1.7% | 4.3% | +2.6 pts | probe.l5.endpoint_live.v1 · 2026-09-30 18:13 UTC |
The ladder
Levels are observations, not judgements. L7 means a wallet bound to the agent paid someone after it registered, with money that left the agent's control; L8 adds that the wallet still holds at least $10 in stablecoins; L9 adds three or more independent counterparties in 30 days. Agents seen paying went from 31 to 5,505 in the period; part of that rise is our payment reader catching up on wallets flagged earlier, which the change register dates.
| Figure | Start of period | End | Change | Definition · measured |
|---|---|---|---|---|
| Agents seen paying (L7+) | 31 | 5,505 | +5,474 | rank.l7plus.v1 · 2026-09-30 21:46 UTC |
| Agents paying and still funded (L8+) | 1 | 440 | +439 | rank.l8plus.v1 · 2026-09-30 21:46 UTC |
| Agents with three or more counterparties (L9) | 0 | 66 | +66 | rank.l9.v1 · 2026-09-30 21:46 UTC |
The money
93,673 counted payments worth $7,654,905 were made by 7,060 agents in September 2026; $4,653,465 of it left the payer's control. 8,576 transfers ($31,648,112) were routing hops and 3,226 ($246,548) were made before the agent registered; neither is counted. 86,217 of the counted payments were read more than three days after they happened, which is our lag, not the market's; the final edition states that number as the last word.
| Figure | Start of period | End | Change | Definition · measured |
|---|---|---|---|---|
| Counted payments | 569 | 79,637 | +79,068 | pay.events.v1 · 2026-09-30 23:27 UTC |
| Counted payments, USD | $66,935 | $25,804,962 | +$25,738,026 | pay.usd.total.v1 · 2026-09-30 23:27 UTC |
| Money that left the agent's control, USD | $3,480,064 | $11,569,912 | +$8,089,849 | pay.usd.leftcontrol.v1 · 2026-09-30 22:50 UTC |
| Deposits into the agent's own positions, USD | $6,017,368 | $14,209,421 | +$8,192,053 | pay.usd.ownpositions.v1 · 2026-09-30 22:50 UTC |
| Addresses paid | 103 | 2,395 | +2,292 | pay.counterparties.v1 · 2026-09-30 23:27 UTC |
| Excluded: routing hops, USD | $1,533,611 | $71,307,966 | +$69,774,355 | pay.excluded.passthrough.v1 · 2026-09-30 23:27 UTC |
| Excluded: before registration, USD | $98,802 | $199,509,543 | +$199,410,741 | pay.excluded.preregistration.v1 · 2026-09-30 21:46 UTC |
| Stablecoins held by bound wallets (sampled), USD | $27,028 | $1,251 | −$25,776 | graph.bound.usdc.v1 · 2026-09-30 18:13 UTC |
| Share of bound wallets funded | 5.0% | 4.5% | -0.5 pts | graph.funded.rate.v1 · 2026-09-30 18:13 UTC |
| Part | USD | Share |
|---|
Where the money went
The fifteen addresses that received the most counted agent money in the period, with our label where we have one. A label is our attribution from public sources, not a statement by the address's owner.
| Address | USD | Payments | Agents |
|---|---|---|---|
| Adapter (Aave v3 / ERC-4626 policy adapter) · lending | $2,912,273 | 1,593 | 465 |
| TermixEscrow · work_escrow | $2,290,768 | 47,734 | 4,247 |
| LI.FI: Permit2Proxy · bridge | $571,581 | 1,345 | 217 |
| LI.FI: LiFiDiamond · bridge | $220,021 | 1,606 | 411 |
| RelayDepository · bridge | $111,567 | 1,015 | 375 |
0xdc2eee55…34f05f | $70,613 | 6 | 1 |
| 0x Settler (BaseSettler) · swap | $57,412 | 535 | 182 |
| KyberSwap executor (MetaAggregationRouterV2) · swap | $57,090 | 274 | 101 |
| Across: SpokePool · bridge | $53,707 | 166 | 51 |
| Compound v3 Comet (CometWithExtendedAssetList) · lending | $50,880 | 47 | 35 |
0x8a311d70…f7422d | $47,411 | 560 | 22 |
| UniswapV3Pool · swap | $46,031 | 271 | 89 |
| OpenRouter (swap router) · swap | $35,767 | 147 | 84 |
0x8b29dabd…40721d | $34,523 | 13 | 4 |
| UniswapV3Pool · swap | $29,730 | 166 | 22 |
The rails
The x402 submitter job read 0 settlements ($0) into 0 Bazaar-listed sellers from 0 payers in the period, 0 of them submitted by a sender that was neither the payer nor a listed facilitator. transactions read by the daily submitter job (a sample of transfers into Bazaar sellers); the stock x402_settle_txs_24h is the count found.
Attacks and exposure
22,121 look-alike addresses were planted into the payment histories of 827 agent wallets; 2 payments ($60) went to a look-alike. 185 detections were logged, 0 of them onboarding moves (a registration handed from the minting account to its holder within a day), which are info only and never live incidents. 0 incidents were live at the end of the period. A detection is a pattern with evidence for a human to read, never a verdict.
| Kind | Severity | Logged | Onboarding moves | Open now |
|---|---|---|---|---|
| uri_timeout_streak | info | 77 | 0 | 0 |
| mint_burst | notable | 56 | 0 | 0 |
| fleet_sibling_exposure | info | 17 | 0 | 0 |
| drain_pattern | notable | 16 | 0 | 0 |
| retry_storm | notable | 12 | 0 | 0 |
| mint_burst | major | 3 | 0 | 0 |
| bind_closed_after_nft_transfer | notable | 2 | 0 | 0 |
| test_then_drain | notable | 1 | 0 | 0 |
| layering_fan | info | 1 | 0 | 0 |
| Figure | Start of period | End | Change | Definition · measured |
|---|---|---|---|---|
| Detections logged, all time | 200 | 428 | +228 | detection.logged.v1 · 2026-09-30 23:51 UTC |
| Live incidents | 1 | 0 | −1 | incident.live.v2 · 2026-09-30 23:51 UTC |
What we changed
2 entries in our change register are dated inside the period, 1 of them corrections of an earlier entry. A step in a series at one of these dates is ours, not the market's; each entry names the series it moves and, where measured, the size of the move.
| Date | Entry | Kind | Summary |
|---|---|---|---|
| 2026-09-27 | cp-2026-09-27-sweep | coverage | Every bound wallet is swept every three hours for USDC, USDT, EURC and DAI transfers (before: a smaller watche…Every bound wallet is swept every three hours for USDC, USDT, EURC and DAI transfers (before: a smaller watched set, USDC only), and the per-wallet scan reads up to 10,000 transfers instead of the newest 50. The counts rose because more of the existing history became visible, not because agents began paying that day: L7+ went 519 → 2,387 on 27 Sep and 7,681 by 3 Oct, when the backfill settled. Compare these series from 3 Oct 2026. |
| 2026-09-27 | cp-2026-09-27-sweep-scopecorrects cp-2026-09-27-sweep | coverage | Scope correction to cp-2026-09-27-sweep: the series computed from the same counted payments (counterparties, p…Scope correction to cp-2026-09-27-sweep: the series computed from the same counted payments (counterparties, payment categories, the x402 rail share, rated L7 actors, idle funded actors) stepped with it between 27 Sep and 3 Oct, while the stored payment rows grew from 15,393 to 277,120. These steps are the backfill, not new activity. Compare these series from 3 Oct 2026. |
Our own checks
The data-quality suite ran 58 times in the period with 2 blocking failures and 71 warnings (37 checks at the end; health 95.4%). 0 statement windows were signed; the log head at the end of the period had n/a leaves. 36 of 37 public feeds were on time when this edition was produced (late: BNB registration owners' payments).
| Check | Severity | Runs failing | First → last |
|---|---|---|---|
actor_evidence_without_rank | warn | 36 | 2026-09-25 → 2026-09-30 |
me_pass_through_backlog | warn | 35 | 2026-09-26 → 2026-09-30 |
me_known_values | block | 1 | 2026-09-27 → 2026-09-27 |
me_pre_registration_leak | block | 1 | 2026-09-27 → 2026-09-27 |
| Series | Day | Before → after |
|---|---|---|
poison_copy_lag_median_min | 2026-10-20 | 18.32 → 140.35 |
approvals_top_spender_wallet_share | 2026-10-19 | 0.70 → 0.49 |
approvals_unlimited_wallets | 2026-10-19 | 619.00 → 376.00 |
poison_watch_wallets | 2026-10-17 | 8,905.00 → 9,343.00 |
solana_agent_owner_outflow_usd_30d | 2026-10-17 | 35,961.64 → 69,474.29 |
treasury_eth_share | 2026-10-13 | 0.50 → 0.44 |
actors_l8_plus | 2026-10-12 | 374.00 → 558.00 |
doppler_distinct_integrators_seen | 2026-10-09 | 1,895.00 → 2,086.00 |
What this edition does not assert
- That a registration is a working agent, or that a level is a judgement of the people behind an agent.
- That counted payments are commerce: deposits into the agent's own positions and bridges back to the payer are inside the counted set and shown separately.
- That any detection is a verdict: each is a pattern with evidence for a human to read.
- That the period's per-day payment figures are complete on the day of production: payments read late are added to the day they were made, and the final edition says how many were read late.
Verify this edition
- The file. Every figure on this page comes from state-of-agent-money-2026-09.json; its SHA-256 is
d0d78f60d546e963b14209dfa71fef108d7c217c2f80eef57f000a6c690a1a78. - The working papers. Every agent, payee and figure behind this edition is a 30¢ pack: buy it on the Desk (sign in with a wallet first), or for an agent
GET /check/edition?period=2026-09. - The log. The final edition's hash is a leaf of our signed statement log:
publication/1|state-of-agent-money-2026-09|d0d78f60d546e963b14209dfa71fef108d7c217c2f80eef57f000a6c690a1a78|….afg logwatchverifies the log's heads and consistency from outside; the leaf list is at /api/statements/log. - The chain. Registrations and payments can be re-counted from any Base node;
afg chaincheck <agent>does it for one agent from a public RPC we do not run, andafg editioncheck 2026-09re-derives this edition's registration figures at the period's own blocks and confirms the file's hash in the log (open-source CLI, 0.3.3). - The definitions. Each figure's definition id resolves at /def, frozen; a change of method is a new version and a dated entry in the register, never a silent edit.
- The PDF. The same edition as a paper: afg-state-of-agent-money-2026-09.pdf, built from the same file.
- Independence. Agentic Finance Graph has no token; AFG and its operator hold no position in the tokens discussed; nobody pays for a level, a rank or a placement; a finding is public before any paid depth of it is sold (stated 24 October 2026, the manifesto).