AI is giving businesses across industries new ways to improve operations and streamline workflows. For customer-facing functions such as contact centers, which are under constant pressure to improve customer experience while controlling costs, the appeal of deploying it is obvious.
But what is less obvious is the security risks that bringing AI into the contact center can entail. Because of their customer-facing nature, contact centers have considerations that back-end AI applications do not.
“Contact center AI gives bad actors new opportunities to go after your data,” says Matthew Kamish, Solutions Executive, Customer Experience at New Era Technology.
Knowing how to implement AI securely is therefore as important as deployment itself. The challenge begins with understanding the environment in which AI is being used: one where customer interactions, sensitive data, and business systems converge.
The More AI Connects, the More Data It Can Reach
Contact center AI does not operate as a standalone tool. Its value often comes from its ability to draw on the information that helps a business understand a customer, resolve an issue, or complete an action.
That often means connecting AI to the wider systems that support the contact center's operations, from CRM platforms to service management and payment applications. And that's part of the problem.
“More data flows increase the number of leakages,” Kamish says.
“With all these API connections, AI can grab something from one platform and surface it in its interface, meaning sensitive information can be pulled if improperly configured.”
Understanding where data is held, how it moves between platforms and what the AI can retrieve at each point is not always straightforward. A single customer interaction may involve multiple systems, APIs and data types, each with its own permissions and unique requirements.
Not only can AI access all this data, but the interface through which that information can be retrieved is one that is exposed to customers.
When the Customer Interaction Becomes the Attack Surface
In a traditional contact center, an attacker would need to compromise a system or persuade an employee to provide access to reach sensitive data. With an AI-powered interface, they can instead try to manipulate the model through the conversation itself.
By carefully constructing a request, an attacker can use prompt injection to influence how something like an AI chatbot interprets its instructions and what information it is willing to return from its connected platforms.
That's because “AI can be influenced in ways traditional software cannot be influenced,” Kamish says.
The same capabilities that allow contact center AI to retrieve account information, answer questions, and help resolve issues can potentially be manipulated to expose information beyond what the customer should be able to access.
This is what makes contact center AI difficult to secure. The same access that allows it to retrieve information, resolve requests and improve service can also be exploited if the AI is manipulated. Simply removing that access may reduce the risk, but it can also limit the value the organization intended AI to deliver.
That makes securing contact center AI more complex than simply limiting access to connected systems.
Balancing Access and Control in Contact Center AI
Contact center AI creates a difficult trade-off. The more information and systems it can draw on, the more useful it can become to customers and agents. But every additional connection can also increase the amount of data that could be exposed if the AI is misconfigured or manipulated.
The cost of getting that balance between utility and security wrong can extend beyond a drag on efficiency. If a customer is exposed to sensitive information, a request is handled incorrectly, or the AI behaves in a way that falls outside the organization's intended service standards, then that can become a customer experience issue.
“If you lose trust with your customer base because you’re not paying attention, you’re going to lose business,” Kamish says.
For organizations operating in regulated sectors, the consequences are even more acute. Sensitive information, payment data, or personal records may be involved in the same interactions AI is intended to make quicker and easier, and should that information be surfaced, the costs will be counted in currency lost to fines in addition to customers lost due to reputational damage.
The question for companies implementing contact center AI is therefore not whether they should. It is how they can realize its service benefits without creating a new and vulnerable path to the systems and information behind every customer interaction.
