Who is behind the agent, and what may it do overall?

Current · 20 October 2026 · With sources and limits

What happened

World (18 October 2026). According to PYMNTS, World, the project co-founded by Sam Altman, launched “World ID for banking and finance”. The aim is to let businesses tell whether an agent moving money or shopping acts for a verified person. In the “AgentKit” toolkit (beta), a person delegates their World ID proofs to an agent, which shows them to websites without revealing personal data. The report does not describe how the scope of the agent’s authority is defined or limited (PYMNTS, 18 Oct 2026).

Proof and Circle (19 October 2026). According to their joint press release, a merchant answers an agent with an identity request (x401) and a USDC payment request (x402). The agent presents a verified credential and a signed approval for that specific purchase; the merchant checks the signature against published keys and keeps signed evidence for disputes. A demo is planned at Money20/20 in Las Vegas. The release says the approval covers the specific purchase and describes no broader limits of authority. All statements come from the companies; we found no independent confirmation (GlobeNewswire, 19 Oct 2026).

Context. On 9 October Biometric Update described the core question shifting from “who is acting?” to “what may it do?” (Biometric Update, 9 Oct 2026). Per Semafor, Amazon and eBay have blocked AI shopping agents; security vendor Cheq reports a hundredfold rise in agent traffic within 45 days (vendor claim) (Semafor, 16 Oct 2026).

Why one scheme does not close the question

Both approaches settle which human consents, which helps merchants who want evidence in a dispute. For the customer, what matters is giving the approval themselves. The agent’s operator wants to be recognisable as genuine. For agents acting for a company there is a further question: which agents belong to this company, and what is each allowed to do beyond a single purchase? A person’s identity or a one-off approval does not answer that by itself. Whether and how the approaches complement each other is open.

What ID MANDAT contributes

ID MANDAT is an open registry: a company declares on its own domain which AI agents act for it and what each may and may not do. Anyone can check this without an account: on the web, through a signed REST API (beta), via MCP or A2A. Revoked or suspended entries are shown as such.

Our stance (not a forecast): the statement “this company has entrusted this agent with these limits” should be publicly checkable without an account, alongside schemes that verify individual people or individual purchases. We do not claim ID MANDAT can be combined with World, Proof or x401/x402, or that merchants use it.

What this does not do

ID MANDAT confirms identity and authority, not creditworthiness, trustworthiness or quality.
Statements come from the company itself, not from an auditor.
It does not verify a natural person or a customer’s individual approval.
It cannot determine whether shops, banks or AI systems accept or prefer entries.
It does not replace payment or liability rules.

Sources

As of: 20 October 2026. The situation may have changed.