Who Pays When Your AI Agent Screws Up? The Checklist Nobody Gave You

Your AI Agent Just Made a $200,000 Mistake. Whose Problem Is It?

A startup now exists specifically to stress-test AI agents for jailbreaks, hallucinations, and data leaks — and insure them. AIUC raised $40M Series A, founded by Anthropic's first product hire, because the market for "what happens when the AI breaks something" is already real.

Courts are moving too. Air Canada lost a chatbot case when its AI promised a refund policy that didn't exist. The ruling was clear: you deployed it, you own what it says.

If you're a small business owner thinking about letting an AI agent handle customer emails, process orders, or access financial data, you need answers to three questions before you turn anything on. Not after.

Question 1: Does Your Vendor Agreement Cap Their Liability?

Read the terms of service for any AI tool you're paying for. Most of them contain a clause that limits the vendor's liability to what you paid them — sometimes to zero.

What that means in practice: if an AI coding agent introduces a bug that crashes your client's system, the vendor's liability might be capped at your $20/month subscription fee. The $200,000 of damage? That's on you.

Action: Before deploying any AI agent that touches customer-facing work or financial systems, search the vendor agreement for "limitation of liability" and "indemnification." Know exactly where the buck stops. If it stops at you, plan accordingly.

Question 2: Does Your Business Insurance Cover AI-Caused Errors?

Most general liability and professional liability policies were written before AI agents existed. Here's the uncomfortable reality: your policy probably doesn't explicitly cover damage caused by an AI system acting on your behalf.

This isn't hypothetical. If an AI chatbot promises a customer a discount your business didn't authorize — exactly the Air Canada scenario — and the customer relies on it, you could be on the hook. Your insurance adjuster may say "that's a technology error, not covered under your general liability."

Action: Call your insurance broker. Ask two questions: (1) "Does my current policy cover errors caused by AI tools I've authorized to act on my business's behalf?" and (2) "If not, what endorsement or policy would close that gap?" Get the answers in writing.

Question 3: What's Your Revert and Recovery Plan?

AI agents don't just answer questions — they take actions. They send emails, modify databases, process payments, update records. When one of those actions goes wrong, you need to undo it fast.

Most businesses using AI have no rollback plan. If an AI agent sends 200 incorrect invoices, most small businesses would be manually sorting it out for days.

Action: Build a one-page recovery plan for every AI agent you deploy. It should answer:

What actions can this agent take? (List every one.)
How fast can you detect a bad action? (Monitoring)
How do you undo each type of action? (Rollback)
Who is responsible for executing the undo? (Accountability)

The Pre-Deployment Checklist

Before any AI agent goes live, fill this out:

AI AGENT PRE-DEPLOYMENT CHECKLIST
=================================
Agent name: _______________
Deployment date: _______________

[ ] VENDOR LIABILITY: I have read the vendor's limitation of liability clause.
    Maximum vendor liability: $________
    My remaining exposure: $________

[ ] INSURANCE: I have confirmed (in writing) whether my business insurance
    covers AI-caused errors.
    Coverage status: ☐ Yes  ☐ No  ☐ Pending broker response
    If no, endorsement needed: _______________

[ ] RECOVERY PLAN: I have documented how to detect, stop, and undo each
    action this agent can take.
    Rollback documented for: _______________
    Monitoring in place for: _______________
    Person responsible for rollback: _______________

[ ] SCOPE: This agent is authorized for the following tasks ONLY:
    1. _______________
    2. _______________
    3. _______________

    This agent is NOT authorized for:
    1. _______________
    2. _______________
    3. _______________

[ ] APPROVAL GATES: The following actions require human approval before
    the agent executes them:
    1. _______________
    2. _______________
    3. _______________

Print it. Fill it in. File it. It takes 30 minutes and it might be the document that saves your business when something goes sideways.

The Bottom Line

AI agents are powerful. They're also liability engines that act on your behalf at machine speed. The businesses that survive the AI adoption curve won't be the ones that deployed fastest — they'll be the ones that knew exactly who was responsible when something went wrong.

Ask the three questions. Fill in the checklist. Then turn on the agent.

Want a fillable PDF version of the checklist? Grab it here — print, complete, and keep on file before any AI deployment.

DEV Community