Author: Prabhjot Singh
Title: Chief Digital Officer
Link: /life-at-iqor/leadership/prabhjot-pj-singh
AI Agents Are Calling Your Contact Center. Who Gave Them Permission?
Pj Singh, Chief Digital Officer
Tom Eggemeier, the CEO of Zendesk, recently told Newsweek that he asked a personal AI agent to renegotiate his cable, streaming, internet and mobile plans. Over four days it contacted those providers 37 times and saved him $3,150. Before that, he hadn't called them in three or four years.
That story tells you where customer service is going. Personal AI agents are going to call, chat and message companies on our behalf, and they will do it far more often than we ever did. The volume is one challenge. The bigger one is trust.
When the “customer” on the line is a bot, how does your team know what it’s allowed to do?
The banks are feeling it first
In October, Axios reported that major U.S. banks were detecting third-party AI agents operating inside customer accounts. In most cases, customers had simply handed their usernames and passwords to an agent the bank hadn’t vetted. In some cases, agents appeared to act on their own, including contacting call centers. Morgan Stanley said it had not seen rogue agents on its platforms. But the response was fast. Morgan Stanley began blocking third-party agents, and JPMorgan, Bank of America and Citi moved to restrict them.
I understand why. Every contact center I’ve worked in was built on one assumption: the person asking for access is the account holder. Authentication was designed to answer a single question. Who are you?
An AI agent changes that. It can have the right password and every security answer. That proves it can get in. It doesn’t prove anyone gave it permission to move money, cancel a service or see sensitive information, and most verification systems were never designed to check.
Blocking buys time. Demand will open the door.
Restricting agents makes sense while authorization is unsolved. It won’t last. Personal agents save people time and money, and customers will keep using them. The banks already see this, which is why many are moving toward read-only access as a first step.
Once authentication and security catch up, I expect banks, service providers and retailers to start accepting verified agent traffic. When they do, volume will follow. If one person’s agent can make 37 contacts in four days, think about that across millions of customers. Contact centers need to be ready before the door opens.
Authentication must verify delegated authority
The next generation of authentication must confirm two things: who the account holder is, and what the agent is allowed to do for them. That changes the questions a contact center must ask:
• Is this a legitimate agent, or is someone using an agent to get into the account?
• Did the customer actually authorize it?
• What is the agent allowed to do, and is that permission still current?
• Does this request reflect what the customer really wants?
Answering those questions takes permission defined by action, approval checkpoints for sensitive requests like payments and cancellations, and an expiration date on every authorization. An agent cleared to check a balance last month should not be able to move money today.
Should a bot answer a bot?
Here is where it gets more interesting. When a customer’s bot talks to a company’s bot, both sides are acting for someone who isn’t in the conversation, and each must trust the other.
It also changes how we route. If an inbound request from a bot doesn’t need human empathy, why send it to a person? It makes more sense to send it to a company bot built to handle bot workflows and authorizations. A balance check or a plan comparison can stay bot to bot. A billing dispute, a hardship situation, or any sign that the customer’s bot misunderstood its instructions should go to a person, with the full context.
That is how we think about it at iQor. Our Agentic iQ workforce handles simple, high-volume interactions and hands off to people, with full context, when a conversation needs human judgment. As bot-to-bot traffic grows, getting that split right will matter even more. And people still set the rules. Trained teams decide what the bots can do and own the decisions that carry real risks for the customer.
Can you show what the agent did, and why?
When an AI agent acts on an account, someone will eventually ask what happened. You need an audit trail that shows who or what made the request, the authorization it carried, the action the company took, and why. That record matters even more when there are bots on both sides.
Visibility helps here too. When you analyze every interaction instead of a sample, you can spot patterns in agent-driven contacts, like repeat attempts or unusual requests that point to a gap in verification.
The agents are already calling
Most verification systems still assume a person is on the line, and few can confirm what an AI agent has been authorized to do. The companies that sort out authorization and routing early will be ready when customers push the door open. And they will.
So here is the question I’d ask every CX leader: if an AI agent called your contact center tomorrow, how would your team know what it’s allowed to do, and should a person or a bot answer?
Talk to iQor about preparing your operation for agent-driven contacts, from smarter routing to the right balance of agentic and human support.
Table of Contents
)
)