Using AI safely in customer support

Support agents read customer personal data in almost every ticket. Tickets contain names, addresses, order histories, and sometimes payment or health details. Much of that should never go into a tool your company has not approved. The risk includes a privacy breach, a regulatory problem, and a loss of the customer’s trust.

Keeping it safe also covers what AI produces and decides. A chatbot that confirms account changes without verifying identity creates a security risk. A drafted reply that discloses another customer’s details creates a breach. Some decisions, such as closing an account or denying a claim, need a person.

Where it comes up

• Summarizing a ticket with personal details

  Remove names, emails, phone numbers, addresses, and payment details before pasting into any tool that is not approved for customer data. Your help desk may include a tool your company has approved for this work.

• Handling an account-access request

  Identity verification follows your team’s process, not an AI draft. AI can help write the reply that explains the steps. It must not decide that a person is who they say they are.

• Reviewing what a chatbot discloses

  Test the chatbot with requests for another person’s order or account details. It should refuse and pass the conversation to a person. Report any case where it does not.

What it looks like at each band

1. E1 · Foundational

   Knows that customer personal data should not go into unapproved tools, but does not always spot it inside a long thread or an attachment. Follows the policy when reminded of it.

2. E2 · Proficient

   Removes personal details before using AI as a habit. Uses approved tools for ticket content. Recognizes decisions that need a person, such as identity verification and account closure.

3. E3 · Distinguished

   Judges risk across a whole workflow, including what data enters each tool, what the chatbot can disclose, and which replies need review. Raises concerns before a new AI feature goes live rather than after. Knows when a ticket should not involve AI at all.

4. E4 · Exceptional

   Shapes the team’s practice: which tools are approved for which data, what chatbots may and may not do, and how customers learn when they are talking to AI where that is required. Handles a mistake openly and fixes the process that allowed it.

Common mistakes

• Pasting a full ticket into a personal account

  Ticket threads carry names, addresses, and order histories in signatures and quoted replies. Remove them, or use the approved tool.

• Letting AI confirm identity

  A model cannot verify that a person owns an account. A convincing message is not proof, and social engineering attempts are written to be convincing.

• Hiding that a reply came from a bot

  Customers who later learn they were talking to an undisclosed bot lose trust in every reply. Follow your company’s disclosure rules, and pass the customer to a person when they ask.

A worked example

Finding the common causes behind a month of shipping-delay complaints.

The full study guide for this skill, for every role, is Keeping it safe. The other five skills for customer support professionals are on the AI skills for customer support professionals page. A one-page summary is in the AI data safety checklist.