What JPMorgan said on the call, and what the confirmation adds

On JPMorgan’s third-quarter earnings call on 13 October, Jamie Dimon stated the bank’s position plainly: “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 added: “You don’t want rogue agents, and it could be thousands of them.”

A JPMorgan spokeswoman then gave CNBC a more operational version: “Protecting our customers is our top priority. Because it is not safe, we do not permit third parties to access accounts by obtaining customers’ banking credentials and logging in on their behalf.”

That second sentence is the most precise wording available, and its edge is narrow. It rules out logging in as the customer; it does not say an AI system may not see your data. CNBC names Meta’s Muse, Spear Street Technology’s Instinct and OpenAI’s dots among the third-party agents at issue.

The Verge relayed the CNBC report the next day, repeating the Dimon quotation and its own paraphrase of the confirmation, and adding no reporting of its own. Every quotation in this article is as printed by CNBC, and the call transcript was not available for comparison.

The block is described as retail, and the detail stops there

CNBC attributes its scope wording to the bank: a JPMorgan spokeswoman confirmed that the bank “currently blocks external AI agents from operating on behalf of retail customers.” Business and commercial accounts go unmentioned. Nothing establishes whether the restriction is US-only or global, and Dimon’s phrase “inside the company” carries no geographic limit. CNBC does not describe how the block is enforced or name any technical or contractual mechanism. Neither report compares JPMorgan’s stance with another bank’s. Our earlier report covered the network side: Amex is building Proof-of-Intent checks on its card network, verifying an agent’s intent at the point of purchase, while JPMorgan refuses at the access side, before a transaction exists.

What still works read-only, and what it cannot do

CNBC, paraphrasing the bank’s position, reports that agents can read some Chase data such as balances when a connector such as Plaid holds the customer’s credentials, but that they cannot send money or pay a bill. The split is CNBC’s gloss: no JPMorgan sentence states it.

The read-only rail is already in commercial use. Plaid’s own post of 8 September says it powers the finance connections in Meta’s Muse, sharing balances, transactions, investment holdings and mortgage information, and that it connects to more than 12,000 US institutions. Plaid describes its role as a permissioned data connection and makes no claim about moving money. One limit is worth keeping: Plaid’s post does not name Chase, so it shows the connector model is live, not that Muse can read Chase balances specifically.

No written policy has surfaced

Neither source cites a document. A check of JPMorgan’s investor relations page found no agent-policy item, and the Chase terms URLs tried returned 404, so no customer agreement text could be read. “JPMorgan says it blocks these agents” is supported; “JPMorgan policy” is not. There is also no timeline, as CNBC reports: “There’s no timeline for when the bank would have safeguards in place to allow agents to directly interact with customer accounts, the spokeswoman added.”

Dimon’s “liability transfer” remark leaves the customer’s position undefined

CNBC prints two sentences from Dimon on liability: “If people are using them, they should understand the risk,” Dimon said. “They should understand there’s a liability transfer.” Nothing in JPMorgan’s wording or in CNBC’s report defines what transfers, to whom, or when. No clause, regulator or worked example is cited, so a customer cannot tell from the public record where they would stand if an agent acted on their account and something went wrong.

The bank’s own pilot moves funds

The sharpest asymmetry in the report concerns JPMorgan’s own tools: “In recent weeks, JPMorgan began a targeted pilot for its own AI tool called Smart Cash, which aims to predict a user’s cash flow and move funds to maximize yield, Chief Financial Officer Jeremy Barnum said Tuesday.” An outside agent may not send money; the bank’s own pilot is built to move it. Dimon said JPMorgan is using AI agents internally and plans to deploy them in customer-facing applications, and CNBC reads the blocking stance as “a defensive move against external agents.” That reading is CNBC’s inference and not a JPMorgan statement, and the report provides no basis for judging the competitive motive either way.

Open banking and consumer choice are not addressed

Neither report mentions open-banking rules, consumer-choice rules or the CFPB, so whether refusing third-party agents sits awkwardly with any of them is unanswered by the available evidence. The CFPB finalised its Section 1033 personal financial data rights rule in October 2024; litigation stayed its deadlines and the agency has since said it will rewrite the rule. That rule concerns consumer-permissioned data access, and nothing in this reporting requires a bank to let an agent act on an account.

What to ask any bank’s agent policy

Five questions follow from this story and apply to other banks as well:

  • Which actions require your own banking credentials, and which can a connector perform with read-only access?
  • Is the restriction limited to consumer accounts, and does it apply everywhere you operate?
  • If an agent acts on your account and the outcome is wrong, where is your liability written down?
  • Do the bank’s own agents operate under the same rules as outside agents?
  • Is there a date for safeguards that would reopen direct agent access?

The liability question is the one to settle before connecting an agent to an account.