When people hear “AI data leak”, they picture an AI company getting hacked. That happens, but it isn’t the common case. The common case is quieter: someone pastes a customer list into a chatbot to tidy it up, uploads a contract to get a summary, or connects an AI note-taker to their mailbox. The data leaves, and nobody in IT ever sees it go.
How company data leaks into AI tools
- Pasting. Customer details, financials, source code or HR notes pasted into a chat window.
- Uploading. Whole documents and spreadsheets dropped in for a summary or analysis.
- Personal accounts. Staff using a personal ChatGPT or Gemini login at work, outside any company agreement.
- Connected apps. AI tools granted access to mail, calendars or shared drives with two clicks, and never removed.
- Browser extensions. AI helpers that can read every page a person opens.
- AI agents. Agents that read and move data on someone’s behalf, often with more access than they need.
What the numbers say
- 57% of employees at large US companies admit entering confidential information into public AI tools such as ChatGPT, Gemini and Copilot. (TELUS Digital survey, January 2025)
- In browser data, 77% of employees who use AI paste data into it, and 82% of those pastes come from personal, unmanaged accounts. (LayerX Enterprise AI and SaaS Data Security Report 2025)
- Shadow AI was involved in 20% of breaches and added about $670,000 to the average breach cost. (IBM Cost of a Data Breach Report 2025)
The pattern is clear: people aren’t being careless on purpose. AI tools are useful, and the safe way to use them hasn’t been made easy.
Why banning AI doesn’t work
Blocking every AI site sounds safe. In practice, people switch to their phones or personal laptops, and you lose sight of the data completely. The companies that do best don’t ban AI. They make the approved tools easy to use, and they catch the risky moments as they happen.
How to stop AI data leaks without stopping AI
- Know what’s in use. Keep a register of every AI tool, who uses it and what it can reach. You can’t govern what you can’t see.
- Pick approved tools and say why. Give people a company-approved option with business terms, so they don’t need personal accounts.
- Write a short AI use policy. One page: what’s allowed, what data never goes into AI, and who to ask. Start from our one-page AI usage policy template.
- Coach in the moment. A short, polite prompt when someone opens an unapproved AI tool works better than an annual training video.
- Close forgotten access. Review AI app connections to mail and drives regularly and remove the ones nobody uses.
- Act when the rules break. When data keeps going out to a tool you don’t allow, block it or revoke its access, and tell the person why.
What MSPs should offer clients
Clients are asking their MSP about AI more than almost anything else. An AI data-leak review is a natural first step: find the AI tools in use, the personal accounts and the open connections, then put a register and simple rules in place. It turns a vague worry into a monthly service clients can see.
How Centeye helps
Centeye Lumen is the AI governance and registry platform. It keeps a register of every AI tool in use and who uses it, shows how much company data goes out to each one (without reading the content), finds personal-account sign-ins and AI apps connected to Microsoft 365 or Google Workspace, and coaches, blocks or revokes on the rules you set. For the AI agents themselves, see Synthwatch and our guide to AI agent governance.
New to the terms? Read about shadow AI, learn how to detect shadow AI, or see AI governance for regulated industries. Want to see what AI tools your clients use? Start a free trial.
Figures cited from the TELUS Digital AI survey (January 2025), the LayerX Enterprise AI and SaaS Data Security Report 2025, and the IBM Cost of a Data Breach Report 2025.