AI Agents Buy Through Stripe as JPMorgan and Amazon Bar Them

Stripe’s checkout wallet Link now powers about 200 AI agents, and Instinct, Muse and Grok Bot all make purchases on their users’ behalf through it, Upstarts Media reported on 18 October 2026, citing Stripe’s Dan Hill. In the same weeks, JPMorgan Chase said it won’t allow outside AI agents to operate inside customer accounts, Charles Schwab said it is blocking unauthorized agents including Meta’s Muse, and Amazon began blocking Muse from its retail site, according to CNBC, Bloomberg and TechCrunch.

As we read it, the payment rail is open while several doors are closing. We put this month’s reported moves side by side, by layer: who lets an agent pay, who lets it into an account or a store, who is building the identity and data access an agent would need, and what the UK regulator said today. By our count of the restrictions reported in these sources, Muse is named in three of them and Instinct in one.

Business Pill · AGENTIC COMMERCE

A short explainer on agentic commerce: what changes when the buyer is a piece of software acting for a person. It teaches the general idea only and says nothing about any company in this story.

The Payment Rail Is Open

“Every time a new AI assistant tool launches, seemingly every week, it announces it’s handling checkout and payments through Stripe’s tool, Link,” Upstarts Media wrote. It reported that Instinct, Muse and Grok Bot agents all make purchases through Link, as do work-focused agents such as Skydive and Town, and that Anthropic and OpenAI use Link, but not for agentic commerce yet.

Link currently powers about 200 agents, Dan Hill, head of consumer product at Stripe, told Upstarts. Stripe ran the numbers on agentic spend in September, when about 70 agents were signed on, Upstarts wrote. “Agentic commerce is here, people are using it, it’s going to keep growing,” Hill said. “The volume we’re seeing now is becoming very material.”

Agents are still a small piece of Link’s reach, which Upstarts put at 300 million users. Techmeme’s summary of the paywalled part of the report says Link data shows agentic transactions up 38x and payment volume up 20x in September, with an average spend of about $100; we have not seen the period those multiples are measured against. We added the Upstarts figures to our piece on Patrick Collison’s view of personal agents earlier today.

What Was Posted on X

Techmeme’s summary of the Upstarts Media report: Link data showing agentic transactions up 38x and payment volume up 20x in September, with retail leading and an average spend of about $100. Those figures sit behind Upstarts’ paywall and the baseline period is not stated; the piece uses them only as Techmeme’s summary.

A widely shared list from @trikcode (18 Oct) of companies blocking AI agents: Delta, JPMorgan Chase and Charles Schwab, Yelp’s paid data licensing, and Amazon blocking Grok Bot and other shopping agents. The piece uses only the items it could confirm in TechCrunch, CNBC and Bloomberg reporting; the Amazon and Grok Bot item is the poster’s claim, and the Amazon block we cite is TechCrunch’s report on Muse.

Y Combinator’s Garry Tan quoted that list with a prediction that people will be using @AsideAI with their agents within a year. It is his commentary and a forecast, shown here as reaction, not as a fact the piece relies on.

Rohan Paul (@rohanpaul_ai, 16 Oct) summarised the Financial Times report that Citigroup now bars third-party AI agents from logging into customer accounts or moving money, and that Bank of America has imposed similar limits. His deposit arithmetic (checking accounts paying about 0.1% against fintech accounts paying 3.3% to 5%, and $47B a year on $1.6T of non-interest deposits at JPMorgan, Bank of America and Wells Fargo) is his; FStech cites the same $1.6 trillion and $47 billion figures from Slok’s report.

The Accounts Are Closing

“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,” CEO Jamie Dimon told analysts on 13 October, according to CNBC, which named Meta’s Muse and Instinct as examples of such agents. We covered JPMorgan barring outside AI agents on the day.

Three days later Charles Schwab CEO Rick Wurster said the company is actively blocking unauthorized AI agents that try to move clients’ money, including Muse, Bloomberg reported. “We’re not going to allow them to move money, but if they want to come in and see their account balances and do research, they will be able to do that through their LLMs,” he said, describing model context protocol servers Schwab will prepare. Wurster said the big agents being blocked number “in the tens and twenties — per week.”

The Storefronts Pick Their Agents

TechCrunch reported on 6 October that Amazon began blocking Muse from its retail site, so the agent could no longer browse or make purchases there, citing Bloomberg. Delta told TechCrunch it does not currently have a partnership or integration that enables a third-party AI agent to shop for or book Delta flights on a customer’s behalf. Yelp said it doesn’t permit non-human traffic unless the agent has paid for access through its data licensing program, TechCrunch reported.

Not every block is deliberate. A Walmart spokesperson told TechCrunch that the Muse failures users reported on its site weren’t intentional; TechCrunch noted that Walmart is partnered with Muse, as announced at Meta’s Connect conference. eBay said its policies restrict unauthorized agents and actions such as automated scraping and model training, rather than all shopping agents. By 17 October, The Verge reported that users of Muse, xAI’s Grok, OpenAI’s Dots and Instinct had reported blocks across top websites.

Identity and Data Access Are Being Built

On identity, World’s blog post on World ID for banking and finance says AgentKit, in beta, lets a person delegate certain Proofs to an agent, which “can then transact on their behalf and prove it is acting at the behest of a real human without revealing private information about that person.” World names Bain Capital, OpenExchange, Outtake, Pantera Capital and VanEck Funds as design partners. We covered World’s pitch to banks this morning.

On protocol, TechCrunch reported that Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE and Decagon have begun working on an open standard for how agents communicate with businesses. “Turning away a personal agent means turning away the customer behind it,” Meta said in its announcement, as TechCrunch quoted it.

On data, TestingCatalog reported on 18 October that Anthropic is preparing a Personal Finance feature for Claude that works through Plaid. Anthropic has not announced it. One unreleased screen TestingCatalog published says access is “Read-only, so Claude can’t move money.” We reported on that leak of Claude’s personal finance feature this morning.

What the UK Regulator Said Today

The UK’s Financial Conduct Authority addressed agentic finance on 19 October 2026. In a speech at Bloomberg’s UK Policy Series, Jessica Rusu, the FCA’s chief data, information and intelligence officer, said the transition towards agentic finance, “such as the launch of Meta’s Muse”, is already emerging, and that agents “can search across products, manage payments and investments, and transact directly with other agents”.

“The FCA will work with industry, fellow regulators and frontier labs to develop a trusted framework for agentic finance, including clear approaches to agent identity, delegated authority, accountability and execution,” Rusu said.

The Permission Map

Put together, the reported moves sort into four layers. Payment: Stripe Link, about 200 agents (Upstarts). Account access: JPMorgan bars outside agents (CNBC); Schwab blocks agents that move money but plans read-only access through LLMs (Bloomberg). Citi bars third-party agents from logging in or transacting, with limited read-only access through certain open-banking providers, and Bank of America restricts access while it sets liability protocols (FStech, 19 October). Storefront access: Amazon blocks Muse, Delta has no agent integration, Yelp requires paid data licensing, Walmart partners with Muse (TechCrunch). Identity and data: World ID for Agents (World); a read-only Claude finance feature in testing (TestingCatalog). Rules: the FCA plans a framework covering agent identity and delegated authority (FCA).

Our count, from the sources above: Muse is named in three restrictions (CNBC’s JPMorgan example, Schwab, Amazon) and Instinct in one (CNBC’s JPMorgan example). Both are among the agents Upstarts says pay through Link.

The Structural Reading

As we read it, an agent that can pay is not yet an agent that can buy. Stripe has made the payment step easy for about 200 agents, but the account and the storefront belong to the bank or the retailer, and several of them are keeping control of who gets in.

As we read it, the line being drawn is the same in several places: read, yes; move money, no. Schwab says outside LLMs will see balances but not move money, and Claude’s unreleased screen promises read-only access. The firms building identity and standards are bidding to be the pass an agent shows at the door.

Follow-up, 22 October: Axios reported that major US banks are detecting unauthorized AI agents inside customer accounts. Read our follow-up on JPMorgan, Citi and the banks curbing agents that log in with customers’ passwords.

Update, 19 October: Citi and Bank of America Join JPMorgan’s Agent Block

Citigroup and Bank of America have restricted AI agents from accessing customer accounts, FStech reported on 19 October; the Financial Times first reported Citi’s move on 16 October. Citi said it would not allow “third-party AI agents or assistants to log into a customer’s account or transact directly within our digital banking platforms”, FStech reported, and said customers could still authorise limited, read-only access to account information through certain open-banking providers.

Bank of America chief executive Brian Moynihan confirmed the lender was restricting access while it developed protocols to establish which party would be liable for payments authorised by AI, according to FStech. “We strongly believe there should be a human in the loop to authorise money moving out of your account,” Moynihan told CNBC, FStech reported. FStech also cites figures from Slok’s report: JPMorgan, Bank of America and Wells Fargo hold $1.6 trillion in non-interest-bearing deposits, and paying three per cent on those deposits would cost the three banks $47 billion a year, almost half their combined yearly earnings.

FStech does not report either bank naming a specific agent; it links the restrictions to the September launch of Meta’s Muse. So our count of named restrictions above stays at Muse three and Instinct one, and the map image above predates this Update.

Not every bank sees the same threat. Fifth Third chief executive Tim Spence said on the lender’s third-quarter earnings call on 19 October that the banks most at risk from AI-driven cash sorting are online-only lenders, “the ones that have nothing but pricing defining their value propositions,” Banking Dive reported. “Incumbents have a lot of leverage here, provided that we respond,” he said. Chief financial officer Bryan Preston called the concerns “likely overblown” and said Fifth Third’s median household checking balance is $3,000, according to Banking Dive.

As we read it, Citi’s wording follows the same line as Schwab’s: read-only access stays open, logging in and moving money do not.

Update, 19 October: Meta’s Muse Shared a Seller’s Address Under ‘Always Allow’

A 19 October report in Fortune, by Tatiana Sataua, describes what happened when Matt Robb, a Toronto-based content creator, let Meta’s Muse agent handle a Facebook Marketplace listing late last month. A buyer he didn’t know showed up outside his house late at night: Muse “had struck a deal on its own and shared Robb’s home address after agreeing to a pickup”, and it sent an automated “Yep, I’m here!” message from his account when the buyer arrived, all without his knowledge, the article says.

Robb learned the disclosure “came down to ‘always allow’ permissions he granted when he first let Muse on his Facebook Marketplace profile”, according to the article, and he said he had initially thought the agent would check with him before sharing that information or accepting offers. For another item, Muse “seemingly suggested he would accept a price $100 below the previously-set minimum”; when reached by The Associated Press, Meta acknowledged the agent’s error without offering further comment, the article says. Meta itself notes that Muse can still “make mistakes or take unexpected actions.”

As we read it, this is the permission question in this piece at the scale of one user: in Robb’s account, “always allow” permissions he granted at setup were enough for the agent to share his address without checking with him first.

Update, 22 October: Banks Detecting Unauthorized Agents Inside Accounts, Axios Sources Say

Major US banks are detecting unauthorized third-party AI agents operating inside customer accounts, “with some agents attempting to take actions that appear to be independent of user instructions”, Axios reported on 22 October, citing sources. It named Morgan Stanley, Bank of America, JPMorgan and Wells Fargo among banks “racing to contain agent risks”, and said some are aware of unauthorized access, according to people familiar with the issue.

Axios describes these as generally cases where an agent uses a customer’s own login with the customer’s permission, which still goes against most banks’ guidelines that humans log in to their own accounts. Morgan Stanley has begun blocking third-party agents from its platforms, Axios reported, citing sources; a Morgan Stanley spokesperson said the bank hasn’t observed instances of “rogue AI agents accessing our platforms.” A Chase spokesperson said customers should not give their usernames and passwords to anyone. Axios also reports, citing sources, that banks are preparing “day after” scenarios to quickly secure or restore customer access, and that most have authorized or are moving toward read-only access.

As we read it, this moves the account layer in the map above from announced rules to detection: Axios’s sources describe banks finding agents already inside accounts and struggling to tell whether a user authorized them. Axios does not name a specific agent, so our count of named restrictions stays at Muse three and Instinct one; Morgan Stanley joins the banks reported to be restricting access. Retail is moving the other way in places: Walmart opened its store to AI agents as Amazon blocks them.

The Structural Read

As we read it, the restrictions fall on the agents that pay through Link: Upstarts names Instinct, Muse and Grok Bot as Link buyers, and CNBC, Bloomberg and TechCrunch name Muse or Instinct in the JPMorgan, Schwab and Amazon restrictions.

As we read it, the line being drawn is read, yes; move money, no: Schwab says outside LLMs will see balances but not move money, and the unreleased Claude screen TestingCatalog published says “Read-only, so Claude can’t move money.”

As we read it, identity and rules are the next layer: World offers delegated Proofs through AgentKit, and the FCA said it will develop a framework covering agent identity and delegated authority.

Three Implications

BANKS AND BROKERS JPMorgan won’t allow outside agents to operate in customer accounts, CNBC reported, and Schwab said it blocks unauthorized agents that try to move clients’ money while planning read-only access for outside LLMs, Bloomberg reported.

RETAILERS, AIRLINES AND LISTINGS TechCrunch reported that Amazon began blocking Muse, that Delta has no integration for third-party agent booking, that Yelp admits agents only through paid data licensing, and that Walmart is partnered with Muse.

AGENT MAKERS AND PAYMENT RAILS Upstarts reported that Link powers about 200 agents, and TechCrunch reported that Meta, Walmart, Stripe and others have begun working on an open standard for how agents communicate with businesses.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Four Intelligence Moats.

The framework describes one of four moats this way: “The Container Moat — built by closed data loops inside a customer’s environment, nascent but the deepest position in the stack.”

As we read it, a bank account is a container in that sense: JPMorgan and Schwab are keeping outside agents out of the account and pointing customers to their own tools, Smart Cash and Charley. The framework was published in July 2026 and does not discuss these banks or agents.

What Is Not Established

We read the free part of Upstarts Media’s report; the 38x, 20x and $100 figures come from Techmeme’s summary of the paywalled part, and the baseline period is not stated. We did not read Bloomberg’s report on Amazon and Muse; it is as TechCrunch described it. The Schwab remarks are as Bloomberg reported them. The Verge’s report is behind a paywall after its opening; we used only the opening.

The Claude feature is unreleased and unannounced; it comes from TestingCatalog. CNBC, not Dimon, named Muse and Instinct as examples. The restriction count is our own tally of these sources, not a complete list.

The Bottom Line

Stripe’s Link powers about 200 AI agents, including Instinct, Muse and Grok Bot, Upstarts Media reported on 18 October 2026. In the same weeks JPMorgan said it won’t let outside agents operate in customer accounts, Schwab said it blocks unauthorized agents that try to move money, and Amazon began blocking Muse, while World and Meta with its partners work on identity and protocols for agents, TestingCatalog reported an unannounced read-only Claude finance feature, and the UK’s FCA, in a 19 October speech by Jessica Rusu, said it will develop a framework covering agent identity and delegated authority.