Why a written AI policy isn't enough
Most small companies start with an "acceptable AI use" policy. It's a good first step, 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, usually without thinking about it.
If an insurer, auditor or client asks "how do you enforce your AI policy?", "we told everyone not to" is not a strong answer.
Why blocking ChatGPT usually backfires
Blocking every AI domain at the firewall or DNS level feels safe, but in practice:
- Staff switch to personal phones or hotspots, so you lose all visibility.
- New AI tools appear weekly, so block lists are always behind.
- Leadership usually wants people using AI, so the block gets quietly removed.
What works: let people use AI, but check what leaves
The approach that holds up is to allow the AI tools you've approved and inspect what gets pasted or uploaded into them. Sensitive content is stopped before it leaves; everything else goes through normally.
- Pick approved tools. Decide which AI tools are allowed (for example a company ChatGPT or Copilot plan) and which are not.
- Inspect prompts and uploads. Check text and files sent to AI sites for things like customer records, patient information, card numbers and API keys.
- Match your real data, not just patterns. Pattern rules ("any 9-digit number") create false alarms that people learn to ignore. Matching against fingerprints of your actual customer or patient records only fires on real leaks.
- Log and review. Keep a record of what was blocked so you can coach people and show auditors the control works.
How Arx Cloud does it
Arx Cloud is a safe-internet service for small teams and MSPs. Your team's devices connect through Arx Cloud, and built-in 1-click AI proxy packs check web and coding AI traffic right away without writing domain lists by hand:
- 1-click AI proxy packs (ChatGPT, Claude, Gemini, Copilot & Cursor): turn on prompt and file-upload checks across web AI chat and coding tools in a single click, stopping HIPAA, PCI & PII records and API secrets before they leave.
- Exact Data Match (Enterprise Hybrid plan): upload a CSV of sensitive HIPAA, PCI & PII records; each value is turned into a one-way keyed fingerprint and the original values are discarded. Only exact matches to your real data are blocked.
- Shadow MCP block & AI Agent Gateway: enable the 1-click Block Unbrokered Shadow MCP proxy rule so laptops cannot connect straight to outside MCP tool servers, and route approved Claude Code, Cursor and Copilot connections (including private internal tools over Arx Private Access) through vaulted keys and human approvals. See how to secure AI agents & MCP servers.
- Web filtering & zero-port office access in the same app, including app-by-app remote access with no open inbound firewall ports.
- Self-service: starts at 5 users ($10/user/mo on a 2-year plan), with a 30-day free trial and no sales call.
Want to see it before signing up? The live demo runs on sample data with no login.