What insurers are actually asking
Wording varies by carrier, but the questions usually fall into three groups:
- Policy: do you have a written policy covering which AI tools staff may use and what data they may share?
- Enforcement: are there technical controls that stop sensitive data from going into unapproved AI tools?
- Visibility: can you see which AI tools are in use and what was blocked?
This isn't legal or insurance advice. Always answer questionnaires accurately and check wording with your broker.
Step 1: Write a short, usable policy
- List approved AI tools (for example, your company's paid ChatGPT, Copilot or Gemini plan).
- Define data that must never be entered: customer or patient records, payment data, credentials, contracts and source code.
- Explain what happens if something is blocked, and who to ask.
- Have staff acknowledge it, and keep the record.
Step 2: Add technical enforcement
Enforcement is where most small businesses get stuck. Options, from lightest to strongest:
- Browser or DNS blocking of unapproved AI sites. Easy, but staff bypass it on personal devices, and it doesn't stop leaks into approved tools.
- In-line data-leak protection that inspects what's pasted or uploaded to AI sites and blocks sensitive content. This is what "enforced" usually means.
- Exact Data Match against your real records, which removes most false alarms.
- AI Agent & MCP action controls if engineers or analysts connect coding agents (Claude Code, Cursor, Copilot) to internal tools: vault upstream API keys, block destructive tools, and require human approval for write actions (MCP setup guide).
Step 3: Keep evidence
Keep the signed policy, a screenshot of your controls, and a monthly export of blocked events. That's what you'll be asked for after an incident.
Where Arx Cloud fits
Arx Cloud gives small teams steps 2 and 3 in one app: AI prompt and upload inspection, web filtering, and an event log you can export or stream to a SIEM. It starts at 5 users with a 30-day free trial.