0
Skip to content

Big Banks Restrict AI Agents From Customer Accounts

ai, tech, machine learningAn American Banker webinar on AI agents that act for customers touched on a concern ACA members share: how agencies can tell when an agent — not a consumer — is on a collection call.

10/30/2026 11:00 A.M.

Customer-facing applications built for humans are now encountering autonomous agents, raising questions for financial services companies about who is actually on the other end of a call.

In an Oct. 29 American Banker webinar, Transmit Security co-founder Mickey Boodaei said consumers are beginning to send AI assistants into their bank accounts — and several of the largest U.S. banks are moving to keep those assistants out.

Boodaei described three stages of AI in customer-facing apps. First, banks automated back-office work such as onboarding approvals and anti-money laundering reviews. Second, they added AI assistants inside their own apps. Banks control both. The third stage is outside their control: assistants built by other companies that act on the customer’s behalf.

“You have zero control,” Boodaei said.

He noted that the newest consumer assistants differ from earlier browser-based tools that require users to watch and intervene. Assistants such as Meta’s Muse take instructions through apps like WhatsApp, open a browser in the cloud, and work around the clock. They can read a consumer’s email and calendar, decide when to log in, and use the same web and mobile screens consumers use.

Where Human-Built Fraud Controls Break

Boodaei said the fraud tools banks have built over two decades assume a person is on the other end. Device identification fails when an agent works from its own cloud browser. Behavioral biometrics have nothing to read because an agent produces no human typing or movement patterns.

He also flagged new threat vectors. Criminals can hijack a customer’s AI assistant — which may have access to email, calendars, and bank accounts — or manipulate it by sending it an email offer it then acts on.

His recommended approach starts with visibility: a bank first needs to identify when a session is agentic and which agent is behind it, then log what the agent does, tie that activity to fraud cases, and set policy on which agents are permitted and what they may do. He cautioned that aggressive bot detection at the network edge — such as through a content delivery network — does not reliably catch agents.

Banks Are Already Acting

This topic has also surfaced in recent news. On JPMorgan Chase’s third-quarter earnings call Oct. 13, CEO Jamie Dimon said the bank would limit outside agents’ access to client accounts, warning, “you don’t want rogue agents, and there could be thousands of them,” according to Banking Dive.

Citi told USA Today it does not permit third-party AI agents to log in to customer accounts or transact directly, though customers may authorize limited read-only access through approved open banking providers.

Bank of America CEO Brian Moynihan said the bank is developing protocols to determine liability for ai-authorized payments, Yahoo Finance reported.

Consumer appetite for fully autonomous agents appears limited for now. A September report from six banks — including Bank of America, Capital One, and NatWest — found that only 24% of consumers were willing to let an AI agent shop and pay on their behalf, according to PYMNTS.

Where ARM Agencies Come In

The webinar focused on banks and did not address debt collection, but ACA International members have raised a related concern: AI agents may act for consumers during collection calls without the agency knowing.

Boodaei noted that banks themselves struggle to determine whether a session with a customer involves a person or an agent. Agencies face a version of that same problem on the phone and in portals.

Because agents can read a customer’s email and decide whether to act, a validation notice or collection letter in a consumer’s inbox could prompt an agent to respond. Kompato forecasts that by late 2026, collectors will routinely encounter AI agents negotiating on consumers’ behalf, noting that consumers already use general-purpose AI tools to draft negotiation scripts and weigh settlement options.

Two questions are worth considering now. First, how can an agency determine that a caller, email, or portal session is an AI agent, and what should staff do when they suspect one?

Second, how do Fair Debt Collection Practices Act and Regulation F rules on communicating with third parties apply when a consumer’s assistant contacts an agency — and was that contact authorized by the consumer?

Kelly Knepper-Stephens, chief legal officer at TrueML, has warned that AI is creating new risks in consumer communications, with bots increasingly contacting agencies on consumers’ behalf.

“This is happening now, and if you haven’t had it happen yet, you should immediately think about how you want it to be handled at your agency,” she said at ACA International’s 2025 Convention & Expo.

Agencies that suspect consumers are using AI assistants to communicate should consider treating authorization as a threshold issue: develop specific protocols for AI-assisted communications and document authorization or consent before engaging in substantive discussions about a debt.

Related Content from ACA:

 

Remember to subscribe to ACA Daily and Member Alerts under your My ACA Assistant profile when logged in to acainternational.org.

If you have executive leadership updates or other member news to share with ACA, contact our communications department at comm@acainternational.org. View our publications page for more information and our news submission guidelines here.

One moment please...

Share Profile

This site utilizes cookies as included in our Privacy Policy and User Agreement.