JPMorgan Shuts Out AI Agents as x401 Offers Them an ID Card

Jeff's Daily Pick. America's biggest bank just locked AI agents out of customer accounts. Two days later, a company launched an ID card for agents. The fight over who your AI can act for has begun.

In the United States, JPMorgan Chase chief executive Jamie Dimon said on the bank's third quarter earnings call on October 13, 2026 that outside AI agents will not be allowed to log in to or act on customer accounts. "We're not going to allow just any outside agent to come in inside the company, because it's just too risky for clients and for JPMorgan," he said, according to CNBC. On October 15, Proof, the company formerly known as Notarize, put x401 into production, an open protocol meant to let websites verify which person an agent represents and what that person approved.

What JPMorgan blocks

Dimon said "you don't want rogue agents, and there could be thousands of them," and spoke of a "liability transfer" when customers authorize outside tools. He added that the bank is not against agents and uses them internally. A spokesperson told CNBC the bank prohibits outside parties from accessing accounts by collecting customers' login credentials, while supporting data sharing through open banking providers. In practice, agents can read limited data such as balances through intermediaries but cannot move money or pay bills, and no timeline was given for "agentic connections." The Financial Times reported that Citi is also restricting agents. Meta's Muse assistant, released on September 8, was named among the tools affected.

What x401 proposes

x401 borrows its name from the HTTP 401 Unauthorized status code, a counterpart to x402, which handles payments. Proof says three pieces are live: an agent identity wallet showing the information and transaction terms requested, the open protocol, and a server for the Model Context Protocol that works with assistants including Meta's Muse, ChatGPT, Grok and Claude. A person is matched to a government ID, reviews the action on their own device, authenticates biometrically and signs the terms; if the agent then changes the amount or recipient, the signature fails. Proof's June announcement cites the ESIGN Act and UETA, which allow contracts formed by electronic agents if the action is "legally attributable to the person to be bound." The specification is a working draft under an Apache 2.0 license.

The two moves describe the same gap from opposite sides: banks cannot tell who is behind an agent, so they refuse it; identity firms want to sell the proof. The next step to watch is whether a large bank accepts signed agent mandates in place of shared passwords.