Proof

Merchants decide which agents get in, and they need a reason to say yes.

During our Money20/20 panel discussion, Visa Intelligent Commerce’s Ines Clark set the bar in two parts. The payment has to stay inside its guardrails, meaning the right category and the right spend amount, and it has to match the intent the person expressed. The agent also has to be verified as a trusted agent before the merchant lets it in.

x401 gives the merchant that check. A site asks the agent for proof of who it represents and a signed approval of the exact terms, and lets in the ones that can show both.

🔗 Watch the full panel discussion: hubs.la/Q04D6cqs0

During our panel discussion at Money20/20, Haley Stelzer-Monahan of @TDBank_US shared what a bank needs before agents get more room: limits, guardrails, and auditability.

TD's corporate clients are already putting agents on invoice matching, reconciliation, forecasting, and chasing down missing remittance data. The agent prepares the payment, with a human still in the loop. As agents take on more of that work, every item on Haley's list becomes a record the bank has to be able to check.

A verified human and a signed mandate are two of those records, and x401 is how a site asks an agent for them.

🔗 Watch the full panel discussion: hubs.la/Q04D635x0

A policy and a mandate are two different things, and @gaganmac drew the line with one example during our Money20/20 panel discussion.

A policy is a standing rule in a wallet: "spend up to $5,000, never more than $100 per transaction." A mandate is the instruction behind a single purchase: "I authorize you to spend up to $250 for getting this service."

Wallets already enforce policies well. In Gagan's words, "I think the explicit mandates are the missing piece."

With x401, the person signs the exact terms of the transaction, and if the amount or the recipient changes, the signature stops verifying. And paired with x402, the agent presents a verified credential and pays in USDC.

🔗 Watch the full panel discussion: hubs.la/Q04CLJMp0

Days before #Money2020, we launched x401, an open protocol that lets any website or API ask an AI agent who it's acting for and what that person approved. It already works with Meta's Muse, ChatGPT, Grok and Claude.

During the event, we hosted an evening at CHICA in The Venetian and put that question to @TDBank_US, Visa Intelligent Commerce, and @Circle, in a panel hosted by @sytaylor.

A bank answers for disputed payments, a network decides which agents get in, and a wallet moves the money. All three need to know which verified person is behind the agent and exactly what that person approved.

Thank you to everyone who joined us! And if you're having FOMO, you can watch the full session on-demand: hubs.la/Q04CPtFk0

#AgenticCommerce #x401

@sytaylor mapped the protocols agents need into four layers: entry, identity, tools and intent. He also points out that x401 can request "the person's signature on specific transaction terms, binding their identity to what they approved."

We wanted to show what that binding looks like when an agent asks a bank to wire $25,000 for a customer, so we drew it out.

On the left is Simon's stack. On the right is the approval the customer sees in their Proof wallet. The identity layer answers who is approving, with a verified legal identity proofed to NIST IAL2. The intent layer answers what they're approving: the amount, the payee and the source account.

The customer approves both with one biometric, and a key on their own device signs them together.

If the agent changes the amount or the payee, the signature no longer checks out. The bank ends up holding one record that ties a verified person to the exact terms, so if the customer later says the agent went rogue, the bank can check.

🔗 Simon's original post: hubs.la/Q04CvRd-0
🔗 If agents are starting to show up in your payment flows, talk to our team: hubs.la/Q04CvRbs0