Note 01 · The Binding Record
Nikolai Simonyan, Agentic Finance Graph · version 2.1, 27 September 2026 · A4
How we join an AI agent's identity to the money it moves, why every number above that join depends on it, and what the first census of ERC-8004 agents on Base shows once the join is made honestly. The note specifies the binding record (methods ranked from a signed challenge down to token ownership, a confidence that is a function of the method, validity by block range, closure on resale), derives the counted set and its two exclusions from it, reports the census, and walks through six real agents by transaction, including one wallet whose $34.6M of pre-registration history would otherwise have been reported as agent spending. It closes with the adversarial cases the record resists, what it makes possible for insurers, platforms, agents and auditors, and the open questions.
Cite as: Simonyan, N. (2026). The Binding Record: how we join an AI agent's identity to the money it moves. Agentic Finance Graph Research Note 01, v2.0, 27 September 2026. https://agenticfinancegraph.com/research/afg-research-note-01-v2-1-the-binding-record-2026-09-27.pdf
Version 1.0 of the same day, a six-page summary of the census, is superseded and remains available. Live values for every figure in the note: the data page and /api/state. Writing about this? The press kit has the numbers with their definitions.
In preparation
- Note 02 · How AI agent wallets get drained. Drain shapes in our own detections, the May 2026 prompt-injection theft, and where policy is actually enforced.
- Note 03 · Two populations. The wallets that pay for machine services over x402 and the identities registered under ERC-8004, measured from the seller side on Base and Solana.
- Note 04 · Agent treasuries. What ranked agents hold, priced, and the flows into lending adapters that dominate counted spending.
To be told when a note is published, connect a wallet on the Desk and tick updates, or write to agenticfinancegraph@proton.me.