The news
AI agents in banking are being allowed to look but not to pay. At Okta's Oktane 2026 conference, Equals Money said customers' AI tools can read its data but cannot instruct payments, SiliconANGLE reported on September 29, 2026.
James Simcox, Equals Money's chief operations and product officer, described the company's Model Context Protocol (MCP) server, a standard connector that lets outside AI tools pull data from a company's systems. Customers can plug AI assistants from their accounting or enterprise resource planning software into it to speed up data exchange, he said. But payment instructions are off limits; access is read-only, with writing allowed only for nonsensitive tasks, according to SiliconANGLE.
Simcox also said the company needs automated monitoring that can spot an agent behaving outside its expected pattern and shut it down immediately. SiliconANGLE reported that Equals Money, as a regulated firm, holds AI agents to the same compliance standards as staff, giving each a separate identity that can be audited and revoked.
Consumer payments are moving faster. In a September 29 blog post, Stripe said agentic purchases made with Link, its digital wallet, increased 38 times over the past month, with agents including Meta's Muse using it. Agents work through what Stripe calls spend requests. Stripe said it plans to add spending controls that let consumers set an agent's budget so it can finish a task without the consumer approving each individual transaction.
Stripe also said Link gives agents consumer-permissioned access to transaction data from more than 12,000 financial institutions covering over 97% of U.S. bank accounts, and that consumers can revoke that access at any time.
Large banks are building in controls from the start. In a September 24 announcement of a five-year partnership with Google Cloud, BNP Paribas (BNP.PA) said each AI agent would be authenticated and granted access only to the resources its assigned task requires, and that agent interactions with its systems would be monitored.
The numbers
- Rise in agentic purchases with Stripe Link over one month (Stripe)
- 38x
- Financial institutions reachable through Link for agent data access (Stripe)
- 12,000+
- Share of U.S. bank accounts covered (Stripe)
- 97%+
- Length of BNP Paribas-Google Cloud partnership
- 5 years
Why CEOs should care
For CFOs and treasurers, the practical question is no longer whether staff will connect AI assistants to bank and payment data, but what those assistants are allowed to do once connected. Write the policy before a vendor's default decides it. A workable starting point mirrors Equals Money: agents may read balances, statements and invoices; any action that moves money, changes a payee or alters bank details needs a person. Revisit the policy only when you can see how an agent's actions are logged and reversed.
For CISOs, treat each agent as a separate identity, the way Equals Money and BNP Paribas describe it. Every agent should have its own credentials, the narrowest access its task needs, a full audit trail and a way to be switched off quickly. Ask vendors who connect to your bank data how they authenticate agents, how you revoke one, and whether they can show you what a specific agent did on a specific day.
For boards and general counsel, the liability map is unwritten. When Stripe's planned budgets let an agent buy without per-purchase approval, the open question is who pays when an agent is tricked or makes a mistake: the consumer, the merchant, the wallet or the agent maker. Companies that let employees use agents with corporate cards face the same question and should settle it in their card and expense policies now.
The bigger picture
The tension shows up across the industry. SiliconANGLE reported on September 28 that Bank of America (BAC) and Capital One (COF) have warned that AI shopping agents could increase scam and fraud risks, and quoted Remitly Global CEO Sebastian Gunningham saying every transaction needs a clear answer on who is accountable. At the same time, payment networks and wallets are racing to make agent checkout easier.
Sibos 2026, a major annual banking conference, is being held in Miami from September 28 to October 1 under the theme of digital finance for AI-driven economies, according to Global Finance. Read-only access is the cautious first step some regulated firms are choosing. How long that caution lasts will depend on whether authentication and liability rules for agents catch up with consumer demand.
What’s next
Watch when Stripe switches on agent budgets, whether Equals Money or other business payment firms open payment instructions to AI tools, and whether card networks or regulators publish rules on who bears losses from agent-made purchases.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
