How to stop employees pasting customer data into ChatGPT
People paste customer data into AI tools to get work done, not to cause harm. Here is what actually reduces it.
By Inveil ·
Summary
In a 2026 Wakefield Research survey for PagerDuty, 34% of office professionals had entered customer data into public AI tools and 66% used AI despite believing policy prohibited it. What works is a short policy plus a check on each device that blocks or redacts sensitive data before it is sent.
Wakefield Research survey of 1,250 office professionals, for PagerDuty
Each dot is 1%
34%
of office professionals have entered customer data into public AI tools.
01
43%
entered work-related correspondence.
02
31%
disclosed financial information or confidential documents.
03
66%
used AI tools despite believing company policy prohibited it.
Figures are as reported by the source linked below.
Why it happens
Nobody pastes a customer list into a chatbot to leak it. They want a summary, a reply or a spreadsheet formula, and the customer data is already in the thing they copied. In the Wakefield Research survey, 43% had entered work-related correspondence into public AI tools and 34% had entered customer data.
Why a ban does not work
66% of respondents used AI tools even though they believed company policy prohibited it. People who are told no keep using AI and stop telling anyone, which leaves you with the same exposure and less visibility.
What actually works
- A short, plain policy: which tools are fine and which data never goes in.
- A check on each device that looks at prompts, files and screenshots before they are sent.
- Block the data that must never leave, such as card numbers. Redact the data the request can do without, such as an email address.
- Tell the person what was removed, so they learn the rule in the moment.
- A record per person and per tool, so you can see what is happening.
Block or redact?
Blocking stops the request, which is right for data that should never leave the company. Redacting strips the sensitive part and lets the rest through, so the person still gets their answer. Most teams want both, set rule by rule.
Where Inveil fits
Inveil does the device check described above. You install one small background app on each machine, choose block or redact for each rule, and see every event in one dashboard. Nothing is routed through a gateway.
Inveil is data loss prevention (DLP) for AI. It checks prompts, files and screenshots on each employee's machine before they reach an AI tool, blocks or redacts sensitive data by your rules, and logs every event.
Common questions
- Can I stop employees using ChatGPT at work altogether?
- You can try, but 66% of respondents in the Wakefield Research survey used AI tools despite believing policy prohibited it. Controlling what data goes in is usually more effective than a ban.
- What counts as customer data in an AI prompt?
- Names, email addresses, phone numbers, account details, SSNs and card numbers, whether typed, pasted, attached or captured in a screenshot.
- Does redacting break the answer the AI gives?
- Usually not. Replacing an email address with [REDACTED] lets the request complete and the reply still arrives.
More blogs
- Smaller companies are more likely to put customer data into public AI tools
- How to stop screenshots with personal info going to AI tools
- What is shadow AI, and how do you stop it?
- What to put in an AI usage policy for a small business
- A Microsoft Purview alternative for small businesses protecting AI use
- AI data protection for small business: where to start
- How to stop employees leaking company data to AI: a practical guide
- How to stop API keys and secrets leaking into Cursor, Copilot and ChatGPT