Will your insurer pay if AI causes the loss? AI and professional indemnity

Assume your professional indemnity policy does not automatically cover an AI-caused mistake. Do not assume it excludes one either. The honest answer, right now, is: check.

The insurance market is moving away from "silent AI", where a policy neither promised nor refused to cover it, towards clearer underwriting questions, exclusions and endorsements. That shift is real and it is happening fast.

But it is not uniform, and it does not automatically hit your professional indemnity (PI) cover the way some of the recent headlines suggest.

So if an AI tool contributes to a client loss, the outcome depends on your wording, your disclosure, and the facts of the claim, not on a single industry-wide switch being flipped.

The safe move is to check your wording and your disclosure duties before you lean on AI for billable work, not after something has already gone wrong.

Two different kinds of cover, and why that matters here

Before going further, it is worth being precise about what has actually changed, because a lot of the recent coverage blurs two different products together.

Commercial general liability (CGL) cover responds to bodily injury, property damage, and personal or advertising injury, the classic "someone got hurt, or something got damaged" claims. Professional indemnity, also called errors-and-omissions (E&O) cover, is different. It responds to professional negligence: mistakes, omissions and bad advice arising from the service you provide. A firm can carry both, but they answer different questions.

The AI development making headlines is a CGL story. It is not, on its own, a professional indemnity story. Conflating the two is where a lot of the current anxiety comes from, and it is worth untangling before you read your own policy.

Silent AI, and why it is ending on the CGL side

For the last couple of years most liability and professional indemnity policies said nothing about AI at all. They neither granted nor excluded it. That silence was comfortable while nobody was testing it, because a policyholder could assume an AI-assisted mistake would be treated like any other. Insurers are now starting to write AI into the wording explicitly, rather than leaving it to chance.

As the law firm Fenwick sets out in its analysis of emerging AI exclusions, insurers are "increasingly seeking to define the boundaries of AI coverage, either through affirmative grants of coverage or, more commonly, through exclusions designed to limit exposure." The clearest concrete example so far is on the CGL side. From January 2026, the US standard-forms body ISO (through its parent Verisk) introduced an optional generative-AI exclusion endorsement, CG 40 47 01 26, for commercial general liability forms. Where an insurer attaches it, the endorsement removes CGL cover for bodily injury, property damage, or personal and advertising injury "arising out of, or attributable to" generative AI.

That is a CGL endorsement, on US forms, and it is optional rather than automatic. It is not a professional indemnity exclusion, and it does not, by itself, change what your PI or E&O policy covers. What it does do is signal direction of travel: when standard forms start carving out AI on one line of cover, other lines tend to follow with their own version, in their own time. Silent AI is not becoming friendlier anywhere. It is being replaced with explicit wording, line by line.

What an AI exclusion actually removes, where one exists

Where PI or E&O policies do carry AI wording, the phrasing tends to be broad. A claim "based upon, arising out of, or attributable to" the use of AI can fall outside cover, and that language is wide enough to catch work where AI played only a supporting part. For a professional firm, that is the uncomfortable bit. If an AI tool drafted the analysis, checked the figures or summarised the research behind a piece of advice, an insurer could argue the resulting claim is AI-attributable even though a person signed it off. In practice, how far that argument goes depends heavily on the exact wording, the facts of the claim, and how a court or ombudsman reads "attributable to" for a mixed claim where AI was one input among several. Broad exclusion language is not always read broadly in practice, but you should not rely on that leniency either.

For E&O and PI cover generally, the sharper risk is subtler than a single exclusion clause. Fenwick describes it as coverage fragmentation and "gap risk", where an AI-related claim falls between traditional lines of cover and no single policy clearly answers for it. There is a second trap specific to professional cover: many policies only respond to your defined "professional services", and if those services were assumed to be performed by qualified people, work substantially produced by a tool may sit outside the definition. The exposure is not always a loud exclusion. Sometimes it is the quiet architecture of the policy.

What may still be covered

It is worth stating the other side plainly, because it gets lost in the noise. AI involvement in a piece of work does not automatically kill your PI cover.

Where the work sits within your covered professional services, your AI use has been disclosed to your insurer, and there is no specific exclusion in play, a professional indemnity policy would generally still be expected to respond. As WTW notes on AI and professional indemnity insurance, PI cover should respond where AI use is declared and falls within the scope of the insured services, subject to the usual policy terms. The risk is concentrated at the edges: undisclosed use, work that strays outside your defined services, or a specific AI exclusion you did not know was there. Get those three things right and most ordinary AI-assisted mistakes should look, to your insurer, like ordinary professional mistakes.

The disclosure question

There is a way to weaken cover you technically have, and it is worth naming. Professional indemnity runs on fair presentation and disclosure. If your insurer asks how you operate and you understate your use of AI, or your proposal form implies work is done a certain way and it is not, a later claim may be challenged for non-disclosure or misrepresentation. The remedy an insurer can pursue depends on the nature of the breach, ranging from a straightforward paid claim through to reduced settlement or, in the more serious cases, a decline. The tool did not even have to cause the problem. The gap between what you told your insurer and how you actually work is what creates the exposure.

So the honest step is to tell your broker how AI genuinely sits in your delivery, before you renew, not to keep it quiet in the hope it never comes up. An insurer that knows and prices your AI use is in a far better position to pay a claim than one that finds out at claim time.

A short disclosure checklist worth working through with your broker:

• Which AI tools you actually use, by name
• Which client-facing use cases they touch
• Whether confidential client data is ever entered into them
• Who reviews AI output before it reaches a client, and how
• Whether clients themselves are told AI is involved
• Whether your AI vendors offer any indemnity or liability protection
• Whether you keep logs or records of AI-assisted work
• Whether staff have had any training on approved use
• Whether your AI use has changed materially since your last renewal

A claim, and how the answer could differ

Picture a straightforward case. A firm uses an AI tool to help draft a report for a client. The tool produces a figure that looks right, a person glances at it under time pressure, and it goes out. The figure is wrong, the client relies on it, and there is a loss. Under an older policy that said nothing about AI, and where the use was properly disclosed and within scope, the firm may have had a stronger argument that this was an ordinary professional error, to be assessed and paid like any other. Under a policy carrying a specific AI exclusion, or where the AI use was never disclosed, the insurer has more room to argue the claim sits outside cover.

Nothing about the firm's intent or care needs to have changed for the outcome to differ. What can change is the language in the policy, and what was or was not said at renewal. That is the shift worth noticing: the same mistake, the same client harm, a potentially different answer from your insurer, depending on wording and disclosure rather than on what actually happened. It is why reading your policy now, while nothing has gone wrong, is the cheap moment to act.

Seven questions for your broker

You do not need to be an insurance expert to protect yourself. You need seven answers in writing.

1. Does our current policy contain an AI or generative-AI exclusion, and how broadly is it worded?
2. Does AI-assisted work still fall within our covered "professional services"?
3. If a third-party AI tool we rely on malfunctions, are we covered, or does that count as someone else's product?
4. What are we required to disclose about our AI use at proposal and renewal?
5. Would the answer change if AI was used by a subcontractor, freelancer or software vendor rather than by us directly?
6. Do we need a cyber, technology E&O, media liability or affirmative AI endorsement alongside our PI cover?
7. Is affirmative AI cover available to us, so that AI work is expressly included rather than argued over later?

Ask them plainly, and keep the replies. A broker who cannot answer these is telling you something useful in itself.

How to stay insurable

The good news is that the discipline which keeps AI safe for clients is largely the same discipline that keeps you insurable. A qualified human signs off on anything that reaches a client. AI use follows a written AI policy that names the approved tools and the hard lines. Every consequential use has an owner, in the spirit of giving each AI agent a clear job description. And you keep a light record of where AI touched the work, so that if a claim ever arises you can show what the tool did and who checked it.

That record does two jobs at once. It is how you stop AI mistakes reaching your clients, and it is also the evidence that makes an insurer comfortable keeping you covered on the terms you expect. Firms that can show controlled, documented, human-supervised AI use are a very different proposition to firms that cannot describe how AI sits in their work at all.

It is worth raising this at your next renewal rather than waiting to be asked. An insurer that has seen your controls, your sign-off process and your records can price the risk sensibly, and is more likely to confirm your AI-assisted work sits inside cover rather than leaving it as an open question. That firm is in a far stronger position than one whose AI use is a silence on both sides of the policy.

The takeaway

Do not assume your policy has your back on AI, and do not assume it has been quietly ripped away either. The market is moving from silence to explicit wording, and that wording is not always drafted in the policyholder's favour. Check whether your AI-assisted work sits inside your covered professional services. Disclose your real use at renewal, in writing. Keep light evidence of human review on anything AI has touched. The firms in the best position when a claim lands are not the ones that avoided AI. They are the ones that can show controlled, documented, supervised use before anything goes wrong.

If you would like help putting that governance in place, so your AI use is documented, supervised and defensible, book a review and we will work through it with you.

---

This is general information, not legal or insurance advice. Check your own cover and obligations with your broker or a qualified professional.