Your AI professional indemnity insurance is about to face a question you may not have considered. Your AI answered a customer question at 11 pm on a Sunday. It got the answer wrong. The customer acted on it, lost money, and now they want you to make it right. Who pays?
Most founders I talk to have never asked their broker whether their cover responds to that scenario. They assume it does, because the AI is just a tool, like a spreadsheet or an email template. That assumption is doing a lot of work it probably can’t support.
This post covers what professional indemnity cover usually says about automated advice, the specific questions to put to your broker, and how to structure your AI setup so the answer is boring instead of expensive. No legal advice here, just the questions worth asking before you need the answer.
Key Takeaways
- AI professional indemnity insurance is rarely a separate product. Most cover sits inside your existing PI policy and depends on how “professional services” is worded.
- Ask your broker one question in writing: does the policy respond if the negligent act was performed by software acting on our behalf?
- Silent AI risk is the real problem. Your policy probably doesn’t mention AI at all, which means the answer is decided at claim time, not now.
- Human-in-the-loop review on any advice, quote or commitment keeps the negligent act attributable to a person, which is what most PI wording contemplates.
- Insurers in NZ, Australia and the UK are adding AI exclusions and AI endorsements to renewals through 2026. Read the renewal schedule, not the summary.
- Log every AI interaction with timestamps and inputs. A claim you can reconstruct is defensible. One you can’t is a settlement.
- The cheapest risk control is scope: let AI handle admin and triage, keep regulated advice and pricing commitments behind a human.
What Professional Indemnity Actually Covers, and Where AI Falls Through
Professional indemnity insurance covers you when a client suffers a financial loss because of a negligent act, error or omission in the professional services you provided. That’s the shape of nearly every PI policy sold in New Zealand and Australia. The wording matters more than the label.
Read your policy schedule, and you’ll usually find “professional services” defined as the services described in your proposal form, performed by you or by your employees, partners or subcontractors. Notice what that list contains. People. Named categories of people.
Software doesn’t appear anywhere in it.
That gap is what brokers have started calling silent AI risk, creating unexpected liability exposure from using generative systems in your day-to-day operations. Your policy neither covers AI-caused loss nor excludes it. It simply doesn’t contemplate it. Which means if a claim lands, the coverage question gets argued by a claims adjuster reading a definition written before your booking bot existed.
Most of the time you’ll probably be fine. If your AI drafted an email that a staff member reviewed and sent, the negligent act is still attributable to a person performing a professional service. The AI was a tool, and PI policies handle tools without difficulty. Nobody argues about whether Excel is covered.
The trouble starts when nobody reviewed anything. An AI receptionist that quotes a price, an automated system that confirms a booking window you can’t meet, a bot that tells a customer their claim qualifies when it doesn’t. In each of those, the professional service was delivered without a human in the chain. That’s the scenario your broker needs to answer for, in writing, before it happens.
Six Questions to Put to Your Broker
Send these by email. You want the answer in a form you can produce later, not a reassuring phone call you half remember.
1. Does our professional services definition include acts performed by software operating without direct human supervision?
This is the whole ballgame. If the answer is no, or the answer is “we’d have to check,” you’ve found the gap. Ask for it to be addressed by endorsement.
2. Does the policy contain any AI, machine learning or automated decision exclusion?
Insurers have been adding these quietly through 2025 and 2026. Some are broad enough to swallow a chatbot. Check the renewal schedule rather than the broker’s summary, because exclusions get added at renewal and summaries get recycled.
3. If we use a third-party AI vendor, whose policy responds first?
Your AI receptionist runs on somebody else’s model, through somebody else’s platform, configured by somebody. The answer determines whether you’re chasing a vendor’s insurer or wearing it yourself. Ask what contractual indemnities your vendor gives you and whether your broker has seen them.
4. Do we have cyber cover, and does it overlap or conflict with PI here?
An AI failure can look like a cyber event, a PI event, or both. Two policies with two insurers arguing about which one responds is a slow, expensive outcome. Ask your broker to map the boundary.
5. What did we disclose about our use of AI at last renewal?
Non-disclosure is the fastest way to lose a claim. If you installed an AI system after your last proposal form and never mentioned it, tell your broker now. Material change in the nature of your operations is exactly the sort of thing insurers expect to hear about.
6. Is there a sublimit?
Some AI endorsements come with a sublimit far below your main indemnity limit. Cover that stops at $250,000 when your limit is $2 million isn’t cover; it’s a gesture.
The Practical Fix Is Scope, Not Paperwork
Here’s the part most risk conversations skip. You can spend three months negotiating policy wording, or you can spend an afternoon deciding what your AI is allowed to do. That is the practical connection between generative AI liability and insurance: reduce the situations where an automated mistake can create a claim, rather than relying on policy wording alone.
I run an AI workforce across my own business. It answers the phone, follows up leads, reactivates dormant contacts, drafts emails, and writes reports. It doesn’t give advice that carries regulatory weight, commit to a price, or confirm a delivery date, because those are the outputs where a wrong answer creates a loss somebody can put a number on.
That split is a risk control your broker will recognise immediately.
The Australian government’s AI Ethics Principles put human oversight and contestability at the centre of responsible deployment. Insurers are reading the same material. When your renewal proposal says “all AI-generated client communications containing advice or pricing are reviewed by a named staff member before release,” you’re describing a control, and controls affect both cover and premium.
Three things make this concrete:
Draw the line at commitment. AI can gather, triage, draft, remind and summarise all day. The moment output becomes a promise the business has to keep, a person signs off. That single rule handles most of the exposure.
Log everything. Every AI interaction, timestamped, with the input it received and the output it produced. When a customer says “your system told me X,” you either have the record, or you have their word against nothing. I’ve watched a two-line log entry end a dispute in ten minutes.
Write it down as policy. One page. What the AI handles, what it escalates, who reviews what, and how often the outputs get spot-checked. Brokers and clients both ask for this now. It also gives your team a clear answer when they wonder whether they should be checking something.
Where This Sits Against the Rest of Your Risk
Insurance is the backstop, not the plan. Cover pays out after a customer has already lost money and already lost confidence in you. The commercial damage arrives well before the claim settles.
Most of the AI incidents I see in small businesses aren’t dramatic. They’re a bot that kept confirming appointments after the calendar broke, or a follow-up sequence that kept messaging someone who’d already complained. Nobody sued. But the business spent a fortnight apologising and lost two clients, which represents the real exposure from using generative AI without clear operational safeguards.
The pattern behind those failures is almost always the same: someone bolted an AI tool onto a process without deciding what happens when the tool is wrong. There was no review step, no escalation path, no log. The tool worked exactly as designed, and the design never accounted for being wrong.
This is why I’d rather talk about the system than the tool. An AI brain that knows your business, paired with an AI workforce that does defined work inside defined boundaries, fails differently than a chatbot someone switched on because it looked useful. The boundaries are the product. That’s also what your broker is asking about when they ask how the AI is governed, even if they use different words.
If you’re mapping this out, the AI automation contract checklist covers what to pin down with a vendor before you sign, and why AI automation fails in small businesses covers the operational versions of these same gaps. Worth reading both before your next renewal conversation.
Wrapping Up
AI professional indemnity insurance isn’t a product you go and buy. It’s a set of questions about wording, disclosure and scope that either get answered now, in an email you keep, or later, by an adjuster who has no reason to be generous.
The six questions above take twenty minutes to send. The scope decision takes an afternoon. Between them, you’ll know whether your cover responds, and you’ll have made the most likely failure scenarios far less likely.
Insurance is worth having. Not needing it is worth more. Build the AI so it can only be wrong about things that don’t cost anybody money, keep a person on anything that becomes a commitment, and log the lot.
Then go and ask your broker anyway.
If you’re putting AI into your business and you’re not sure where the risk sits, book a 30-minute Discovery Call. We’ll go through what you’re running, where a wrong answer would actually hurt, and what needs a human on it. No pitch, just a straight conversation about your setup.
Frequently Asked Questions
Does professional indemnity insurance cover AI mistakes?
Usually it depends on whether a person was in the chain. If your AI drafted something a staff member reviewed and approved, the negligent act is attributable to a person providing a professional service, and most policies respond normally. If the AI acted alone, the coverage question turns on how “professional services” is defined in your policy. Ask your broker to confirm in writing.
Do I need separate AI insurance for my business?
Standalone AI liability products exist but are mostly aimed at larger organisations and AI developers. For a small business, the more useful step is getting your existing professional indemnity and cyber policies checked for AI gaps and adding an endorsement if needed. That costs far less than a new policy and addresses the same exposure.
What is silent AI risk?
Silent AI risk describes a policy that neither covers nor excludes AI-caused loss, because the wording predates the technology. The exposure isn’t that you’re uninsured; it’s that nobody knows until a claim is made. Insurers are progressively resolving this by adding either explicit AI exclusions or explicit AI endorsements at renewal.
Who is liable if an AI chatbot gives a customer wrong information?
In most cases, the business that deployed the chatbot is liable, because the customer dealt with you, not your vendor. Courts and regulators have generally treated the deploying business as responsible for what its automated systems tell customers. Vendor contracts may give you a right to recover, but that’s a separate argument you have after you’ve already paid the customer.
Do I have to tell my insurer that I use AI?
Yes, if it represents a material change to how you deliver services. Non-disclosure is one of the most common reasons claims get declined. If you installed an AI receptionist, chatbot or automated advice system after your last proposal form, tell your broker before renewal rather than after an incident. It’s a short email that protects the whole policy.
How do I reduce AI liability risk without slowing the business down?
Keep AI on gathering, triage, drafting and follow-up, and put a human on anything that becomes a commitment: advice with regulatory weight, pricing, and delivery promises. Log every interaction with timestamps and inputs. Write a one-page policy covering what the AI handles and what it escalates. None of that adds friction to the work AI is genuinely good at.
Will using AI increase my professional indemnity premium?
Not automatically. Insurers price risk controls, so a business that can describe its review steps, escalation rules and logging often prices better than one running AI with no documented oversight. What tends to raise premiums or trigger exclusions is unsupervised automation in advice-giving or regulated activity. Bring the controls to the renewal conversation.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.