FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Start with the liability position, because it is the part firms most often get wrong. Your client’s contract is with your firm. If a report, a set of accounts, a valuation or a piece of advice contains an error, the client claims against you — it makes no difference whether the error originated in a spreadsheet, a junior’s draft or a generative AI tool. The professional duty to exercise reasonable skill and care is yours and cannot be delegated to software.
Recovering from the AI vendor afterwards is rarely realistic. Most AI tools are supplied on standard terms that disclaim accuracy, exclude consequential loss and cap liability at trivial sums, often the subscription fee. So the practical position is simple: the firm carries the claim, and the firm’s PI policy is the thing that stands behind it — if the wording responds.
Most PI policies in force today say nothing about artificial intelligence at all. Insurers call this “silent AI”: the risk is neither expressly covered nor expressly excluded, so a claim involving AI-assisted work falls to be decided under general policy language written for a different era.
Silence often works in the policyholder’s favour — a negligent piece of advice is a negligent piece of advice, however it was produced. But silence also leaves room for argument at exactly the moment you need certainty: was the work “professional services” as defined? Was using the tool a fact the insurer should have been told? Ambiguity in a policy is never your friend once a claim is running, and the market is now moving to eliminate it in both directions — some wordings adding express AI exclusions, others adding affirmative AI cover. That is precisely why the wording, not the marketing, is what needs reading.
The scenarios that worry underwriters are not science fiction. They are ordinary professional failures with an AI accelerant: a fabricated citation or invented figure passed into client work unchecked; a drafting error replicated across dozens of documents at machine speed; advice built on a confident-sounding summary that was simply wrong. In each case the negligence is human — the failure to review — but the tool changes the scale and speed at which one lapse becomes many claims.
That is why human review is the heart of the risk story. A firm that can show every AI-assisted output passes through qualified human checking before it reaches a client is presenting a familiar, insurable risk. A firm that cannot is presenting something an underwriter will struggle to price — and a claim a defence lawyer will struggle to run.
Under the Insurance Act 2015, a commercial policyholder owes a duty of fair presentation: you must disclose what you know, or ought to know, that would influence a prudent insurer. Underwriters increasingly ask directly whether firms use AI in client work, and proposal forms are beginning to carry specific questions about it. If the form asks and the answer given is wrong, or if material AI use is simply never mentioned, the insurer may have remedies that reduce or defeat a claim later.
The answer is not to hide the tools — it is to present them properly: what you use, what it is used for, and the review and quality-assurance process around it. Presented that way, AI use is routinely insurable. Discovered later, it is a coverage argument waiting to happen.
Three practical steps. First, read your current wording — or have your broker read it — specifically for AI exclusions, technology definitions and anything that turns on how work is produced. Second, write down how AI is actually used across the firm, including the unofficial uses, because that is what a fair presentation has to cover. Third, put the human-review process on paper. Those three documents turn an awkward renewal conversation into a straightforward one, and they are exactly what we build in a PI programme review.
Your firm is. The client’s contract and your professional duty of care sit with the firm, not the software vendor, and vendor terms typically cap the vendor’s own liability at very low levels. The realistic route to recovery for your client is a claim against you, which is why your PI wording matters.
Some now do, most do not — yet. The majority of wordings in force were written before generative AI and are simply silent. A smaller number of policies have begun adding express AI exclusions, and a few offer affirmative AI cover. The only way to know where yours sits is to read the wording.
If AI use is material to the risk — and use in client-facing work generally is — the duty of fair presentation under the Insurance Act 2015 points firmly towards disclosing it, and many proposal forms now ask directly. Answering inaccurately, or staying quiet about material use, can give the insurer remedies that cut across a later claim.
A policy that neither mentions nor excludes AI. Claims involving AI-assisted work are then decided under general wording written before the technology existed. Silence can favour the policyholder, but it invites argument, and insurers are steadily replacing it with express exclusions or express cover.
Apex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority (FRN 724952). This page is general information, not advice on a specific policy.
