Someone on your team has probably already pasted a customer list, a patient note, or an API key into a public AI tool. Here is what actually works for teams of 5 to 250 people, and what does not.

A written policy is a start, not a fix

Most small companies begin with an acceptable use policy for AI, and cyber insurers increasingly ask for one. But a policy only describes what people should do. Under a deadline, someone will paste the spreadsheet anyway. If an insurer or client asks how you enforce it, "we told everyone not to" is not a strong answer.

Blocking ChatGPT usually backfires

  • Staff switch to personal phones, so you lose all visibility.
  • New AI tools show up every week, so block lists are always behind.
  • Leadership usually wants people using AI, so the block quietly gets removed.

What works: let people use AI, but check what leaves

  1. Pick approved tools. Decide which AI tools are allowed, for example a company ChatGPT or Copilot plan.
  2. Check pastes and uploads. Look at text and files going to AI sites for customer records, patient information, card numbers, and API keys.
  3. Match your real data, not just patterns. A rule like "any 9-digit number" creates false alarms that people learn to ignore. Matching against your actual records only fires on real leaks.
  4. Keep a log. A record of what was blocked lets you coach people and show auditors the control works.

Tools that do this well are usually sold on enterprise contracts with large seat minimums, which most 10 to 100 person companies cannot buy.

How Arx Cloud does it

  • One click turns on paste and upload checks for ChatGPT, Claude, Gemini, Copilot, and Cursor.
  • On the higher business plan you can upload a list of your real customer or patient records. Only real matches are stopped, and the original values are not kept.
  • Web filtering and private office access are in the same app.
  • It starts at 5 users, is self-service, and includes a 30-day free trial with no sales call.

Quick answers

Can staff keep using ChatGPT? Yes. That is the point. Approved tools keep working and only sensitive content is stopped.

Does Arx Cloud store our sensitive records? No. Each value is turned into a one-way fingerprint and the original is discarded.

Originally published at https://arxcloud.io/use-cases/stop-chatgpt-data-leaks/. Side-by-side breakdowns: Arx Cloud vs. Netskope & Proofpoint and Arx Cloud vs. Tailscale & NordVPN.