Ask a professional liability underwriter what AI has cost accounting firms in claims so far and the answer is close to nothing.
What Gets Priced When There Is Nothing to Price
Stan Sterna, senior vice president and risk control lead at Aon, put it plainly in a September piece in Accounting Today: there has been no substantive claim activity linked to the use of AI at accounting firms, and it takes years to understand a new risk. Not many claims, not much paid.
Read that as good news for about five seconds, then look at what insurers are doing in the absence of loss data.
Underwriters cannot wait for a decade of claims before writing next year's policy. So they substitute the thing they can observe.
Sterna describes the current line of questioning as three questions. Do you use AI. Do you police it. Do you have protocols in place. He expects more detailed questions and guidelines within one to two years, most likely built around human review as the primary control.
Gary Florian, senior vice president of underwriting at Camico, frames the same point from the carrier side: the concern is not whether a firm uses AI but how, with data security central and results treated as something to verify before relying on them.
None of that is a loss ratio. It is a proxy, and the proxy is documentation. A firm that can describe which tools it uses, on what work, with what review step, is answering the question. A firm that says everyone uses it a bit for drafting is describing an unmeasured exposure, and unmeasured exposures get priced conservatively.
The Exclusions Are Already Moving Elsewhere
There is a second thing happening on adjacent paper, and it is worth watching because it shows the direction.
In January 2026 the Insurance Services Office introduced a generative AI exclusion for commercial general liability policies, cutting coverage for bodily injury, property damage and personal or advertising injury arising out of or attributable to generative AI. Legal commentary tracking the shift reports AI exclusions and endorsements now appearing across directors and officers, employment practices, fiduciary and technology errors and omissions policies too.
Accountants' professional liability has not gone that way. It is still in the questionnaire phase. But the gap between a market that asks questions and a market that writes exclusions has historically been a few renewal cycles, and it closes faster when the first large claim lands.
The Cheapest Move Is in the Engagement Letter
John Raspante, director of risk management at McGowan, recommends something small enough to do this month: put AI disclosure language in the engagement letter and give the client an opt-out. He also suggests a strong compliance programme around AI might reduce premiums rather than merely satisfy a form.
That is an unusual position for a risk manager to take, and it is worth taking seriously for a reason beyond insurance. A client who learns from the engagement letter that AI may be used on their work, and who is offered the choice, cannot later say they were not told. A client who finds out some other way has a grievance that did not need to exist.
What the Evidence Actually Has to Be
The gap between a firm that has a policy and a firm that can evidence one is where this gets uncomfortable, because the second is a system question rather than a writing exercise.
A policy document says what should happen. An underwriter asking whether you police it is asking what did happen, and answering that means records: which tools were used, on whose work, who reviewed the output, who could see the client data and when.
MetaWurks is built with that side in mind. Documents ingested into the platform are not used to train models and are not exposed to other users, role based access controls decide who can open which client's records, and audit logs record who opened what and when. Those logs are unremarkable until somebody asks a question about March, at which point they are the difference between a record and a recollection.
It is not a governance programme and it does not fill in a renewal form. What it does is make the firm's stated practice checkable, which is precisely what the questions are reaching for.
No claims yet is a window, not a verdict. The firms that use it will renew as a documented risk. The rest will renew as an unknown one, and unknown is the more expensive category in every insurance market there has ever been.
Join the Conversation
If your carrier asked today which AI tools touched client work this quarter and who reviewed the output, could your firm answer from records rather than memory?