Most contact centre security planning treats AI as a future decision: should we adopt it, and if so, how do we secure it. That framing misses what has already happened. While leadership debates the question, agents on the floor are already using AI tools the company never approved, and feeding them customer data to get through the day.
This is shadow AI, and in a contact centre it is not a risk to prevent. It is an incident to find. This article explains why, and how to look.
What shadow AI looks like on the floor
Shadow AI is any AI tool used for work that the organisation did not sanction or secure. In a contact centre it is mundane and everywhere. An agent pastes a customer's angry message into a public chatbot to get help drafting a calm reply. Another uses an AI tool to summarise a long case history. A team lead runs customer data through an AI spreadsheet assistant to spot a pattern.
None of these agents is acting maliciously. They are doing what the tool makes easy, to handle their workload. But each of those actions sends customer data, sometimes including names, account details, or complaint content, into a system outside the company's control, with no record that it happened.
Why the contact centre is where this lands first
Kevin Davis, who writes KD Be Schemin, has pointed to the scale of the problem: a large majority of organisations report AI-related security incidents, while only a small minority govern AI tools as identity-bearing systems with their own access and accountability. The gap between those two numbers is the shadow AI problem.
The contact centre is where it concentrates, for a simple reason. Customer service agents are under constant time pressure, they handle customer data all day, and they are resourceful. Give a resourceful, time-pressured person a tool that makes the job faster and they will use it, approved or not. The contact centre is the part of the company most likely to adopt unmanaged AI fastest, and the part holding the most customer data while it does.
This is the shadow IT story, repeating
The pattern is not new. A decade ago, staff adopted unsanctioned cloud apps faster than IT could approve them, and the term shadow IT was coined for it. The lesson then was that banning the tools did not work; people used them anyway, just more quietly. What worked was finding what was already in use and bringing it under management.
Shadow AI is the same story with higher stakes, because the data going into these tools is customer data and the tools can act on it. Treating it as a future adoption decision repeats the original mistake. The decision was already made, on the floor, weeks ago.
How to find it
Because shadow AI leaves little trace, you have to go looking rather than wait for an alert. The search is part technical and part human.
On the technical side, network and endpoint logs can show which AI services contact-centre machines are reaching. On the human side, ask the agents directly, without blame: which AI tools make your job easier, and what do you put into them. Agents will usually tell you, because they do not think of it as wrong. Nick Clark, who writes Service Matters, has argued that the contact centre is the security weak link precisely because frontline reality runs ahead of policy. The way to close the gap is to see the reality first.
From incident to managed tool
Finding shadow AI is not about discipline. An agent using a chatbot to draft a better reply is trying to do the job well. The point of the audit is to convert that genuine need into a sanctioned, secured tool: give agents an approved AI assistant that does the same job without sending customer data into the open.
The security gap in your contact centre is not a decision still ahead of you. Staff are already using unsanctioned AI tools with customer data. The work is to find what is in use and bring it under management.