Malpractice insurance and AI: the questions carriers now ask law firms

Are malpractice carriers really excluding AI-related claims?

The short answer, based on what has been publicly reported, is not yet in the standard lawyers-professional-liability market. As of mid-2026, industry reporting does not identify any major U.S. LPL carrier that has attached a named AI exclusion to its standard malpractice policies. An AI-assisted error, such as an unverified citation that reaches a filing, is generally analyzed under the existing terms of the policy like any other professional mistake. That is a meaningfully calmer picture than some marketing content suggests, and it is worth stating plainly before discussing where the market is moving.

It helps to understand why that is the case. A professional-liability policy is built around the concept of a negligent act, error, or omission in the course of providing legal services. When an attorney files a brief containing a fabricated citation, the tool that produced the draft is, from the policy's standpoint, part of the how rather than a separate category of harm. The lawyer remains responsible for the work product, and the policy continues to respond to the professional error the way it always has. Naming a specific technology in an exclusion is a deliberate act by an insurer, and the reporting available does not show major LPL carriers taking that step in their standard forms today.

The movement is real, though, in adjacent lines. Pillsbury's Policyholder Pulse reported in April 2026 that AI exclusions are appearing in insurance policies with broad language and uncertain impact, and urged policyholders to scrutinize the wording carefully ^[1]. Verisk, whose standard forms shape much of the market, has rolled out new general-liability exclusions addressing generative-AI exposures ^[2]. Neither development rewrites LPL coverage today. Both signal the direction underwriters are thinking, which is exactly why the questions on your renewal application are changing first.

This article describes reported market conditions as of mid-2026. It is general information, not legal or insurance advice, and it does not interpret the terms of any specific policy. Read your own policy and endorsements with your broker.

Why the questions change before the coverage does

Questions are the leading indicator. Exclusions, when they come, are the lagging one. A firm that treats today's application questions seriously is preparing for tomorrow's policy language.

Insurance markets tend to respond to a new exposure in a predictable order. First underwriters gather information, because they cannot price what they cannot see. Then, as claims data accumulates and patterns become legible, policy language may follow. Exclusions and endorsements are downstream of understanding, and understanding starts with questions. That is why the earliest and most visible sign of AI's effect on legal malpractice coverage is not a change to the policy form but a change to the application.

This sequencing matters for how a firm should read the current moment. The absence of a named AI exclusion in standard LPL policies is not a signal that carriers are indifferent to the risk. It is a signal that they are still in the information-gathering phase for this line, even as the general-liability market has already begun to name the exposure in its forms ^[2]. A managing partner who waits for an exclusion to appear before taking AI governance seriously has misread the order of events. The underwriting conversation is happening now, through the questionnaire, and the answers a firm gives today become part of the record a carrier relies on later.

There is a second reason the questions arrive first. An exclusion is a blunt instrument that reduces the value of the coverage for everyone, and carriers are generally reluctant to devalue a product while they still have room to differentiate between applicants. Questions let an underwriter separate the well-governed firm from the poorly governed one and price accordingly. For a firm with real controls, that differentiation is an opportunity rather than a threat, because it rewards the work the firm has already done.

The AI questions appearing on renewal applications

Before exclusions arrive in a line of coverage, questions do. Underwriters price risk with information, and generative AI is a risk they currently understand mostly through what applicants tell them. Industry reporting indicates that LPL underwriters are increasingly asking AI-governance questions at renewal, even while their policy forms remain unchanged. The questions vary by carrier, but they cluster around a recognizable set of themes that any managing partner can prepare for:

Notice what these questions have in common: none of them ask whether you use AI. They ask whether you govern it. A firm that uses AI heavily under documented controls will generally read better on paper than a firm that claims not to use AI at all while associates quietly run consumer chatbots on client matters. The application is probing for the gap between stated policy and actual practice, because that gap is where malpractice claims are born.

Read the list carefully and a structure emerges. Some questions are about scope, meaning what tools touch what work. Some are about data, meaning where client information travels. Some are about verification, meaning what human check stands between an AI output and a filing. And some are about governance maturity, meaning whether the firm has written the rules down and trained its people on them. An underwriter is not looking for a single perfect answer to any one of these. They are looking for internal consistency across all of them, because a firm whose answers hang together is a firm that has actually thought about the problem.

• Which generative AI tools the firm uses, and for what categories of work
• Whether client data enters third-party models, and under what safeguards
• Whether AI outputs, including citations, are verified before anything is filed
• Whether the firm has a written AI policy and provides training on it
• How AI incidents would be detected, escalated, and documented

What non-disclosure at renewal can cost you

The most dangerous AI risk at renewal is not the technology. It is an inaccurate answer about the technology. Insurance applications are the factual foundation of the policy, and material misstatements or omissions can give a carrier grounds to contest coverage when a claim arrives. If a firm answers that AI outputs are verified before filing, and a later claim file shows a fabricated citation went out unchecked, the firm faces two problems at once: the underlying malpractice allegation and a coverage dispute over what it told its insurer.

The compounding nature of that second problem is what makes it so costly. A single professional error is a manageable event, the kind of thing malpractice coverage exists to absorb. But when the error sits next to an application answer that the claim file contradicts, the firm has effectively handed its carrier a reason to look harder at the disclosure itself. What could have been a straightforward claim becomes a two-front dispute, and the firm defends both its work and its paperwork at the same time. The inaccurate answer did not cause the underlying mistake, but it removed the cushion the firm was paying to have.

This is why the honest answer, carefully documented, beats the flattering one. If your verification process is informal, say so and describe it accurately, then improve it before the next renewal. Involve your broker early: brokers see how different carriers frame AI questions and can help you answer precisely rather than expansively. And treat the renewal application as a governance audit with a deadline. Every question a carrier asks about AI is a question a plaintiff's lawyer could ask in discovery later, and the firm that has real answers for one has real answers for both.

How to prepare accurate answers before renewal season

Accuracy on an application is not a matter of finding the right words at the last minute. It is a matter of knowing, before the questionnaire arrives, what is actually true about how your firm uses AI. That knowledge is built through a short, deliberate process rather than a scramble, and the process is worth running well ahead of the renewal date so that any gaps it surfaces can be fixed before an underwriter reads about them.

The point of preparing early is not to make the firm look better than it is. It is to make the firm's answers match its practices, and to give it a window to close the distance between the two where that distance is uncomfortable. A firm that discovers in this process that its verification is informal has learned something useful, and it has time to formalize the process and describe it honestly at renewal rather than either overstating it or hiding it.

1. Inventory actual usageFind out which tools the firm and its people really use, on which categories of matters, rather than which tools the policy says are allowed. The gap between the two is the first thing worth knowing.
2. Trace where client data goesEstablish whether any client information reaches third-party models, and under what safeguards. This is one of the questions carriers ask most directly, so answer it for yourself first.
3. Describe verification honestlyWrite down how AI outputs, including citations, are checked before a filing goes out. If the process is informal, record it as informal rather than dressing it up.
4. Close the gaps you canWhere the honest description is weaker than you want it to be, improve the underlying control before renewal and keep a record of the change. Then the accurate answer and the better answer become the same answer.

The documentation that makes a firm a better AI risk on paper

Underwriting rewards demonstrable controls. When a carrier asks whether AI outputs are verified before filing, there is a wide quality gap between the answer "yes, our attorneys are careful" and the answer "yes, and here is a signed, independently verifiable record of the citation certification run on every filing this year." The first is an assurance. The second is evidence. Firms in every other risk domain have learned this lesson: documented controls, consistently applied and independently checkable, are what turn a questionnaire answer into a credible underwriting fact.

The difference is not cosmetic. An assurance asks the underwriter to trust the applicant's characterization of itself, and underwriters know that every applicant characterizes itself favorably. Evidence asks the underwriter to trust a record instead, one that exists independently of how the firm chooses to describe it. A record that was created contemporaneously, that cannot be quietly revised after the fact, and that a third party could check carries a weight that no adjective can match. This is why regulated industries lean so heavily on documentation: it converts a claim about behavior into an artifact of behavior.

This is the role RankShield Legal plays, stated honestly. The platform produces signed records of certified citation checks before filings and attested records that privileged material never reached third-party models, so a firm can answer carrier questions with verifiable documentation instead of assurances. What that documentation cannot do is set your price. Premiums and coverage decisions belong to carriers, and no vendor, RankShield included, can promise that governance evidence lowers a quote. The defensible claim is narrower and more useful: verifiable records make a firm a better risk on paper, and better paper is what underwriters actually read.

Assurances versus evidence: what underwriters actually read

It is worth dwelling on the distinction between an assurance and a piece of evidence, because it is the hinge on which the whole application turns. An assurance is a statement about the future or about a general disposition. "Our attorneys verify their work" is an assurance. It may be entirely sincere, and it may even be accurate most of the time, but it points to an intention rather than a fact that can be examined. When a claim arrives, an assurance offers the firm very little, because the question is no longer what the firm intended to do but what it actually did on the matter in dispute.

Evidence points backward at specific events. A signed record of a citation check performed before a particular filing is a fact about that filing, not a disposition about the firm. It survives the moment it was made, and it can be produced later without depending on anyone's memory. That durability is exactly what makes it valuable both to an underwriter deciding how to price the firm and to the firm itself if it ever has to reconstruct what happened on a matter that went sideways. The same record answers the carrier's question at renewal and the plaintiff's question in discovery.

None of this is unique to AI. It is the ordinary logic of risk transfer, which has always favored the party that can show its work. What is new is that generative AI has created a category of professional activity that firms are only beginning to document, and that carriers are only beginning to ask about. The firms that get ahead of that documentation are not doing anything exotic. They are applying an old discipline to a new tool.

Standalone AI endorsements: what to watch as the market matures

The likeliest near-term future is not a sudden wave of AI exclusions on LPL policies. It is a gradual differentiation: more detailed application questions, then endorsements that clarify or restrict how AI-related errors are treated, and possibly standalone AI endorsements or affirmative-coverage products as carriers accumulate claims data. The general-liability market offers a preview, with Verisk's new generative-AI exclusions showing how standard-form language can move once an exposure is named ^[2]. LPL tends to move more slowly, but it watches the same signals.

If that path holds, the differentiation phase is the one that matters most to firms, because it is the phase in which good governance is rewarded rather than merely noted. When carriers have enough data to separate applicants by their AI controls, the questionnaire stops being a formality and starts influencing terms. A firm that has spent the intervening renewals building a documented record of its controls will enter that phase with something to show, while a firm that treated the questions as boilerplate will enter it with assurances and hope. The distance between those two positions widens with every renewal that passes.

The practical takeaway is to read wording, not headlines. Pillsbury's analysis of early AI exclusions stresses that the language is broad and its impact uncertain, which cuts both ways: a vaguely worded exclusion could reach further than intended, and a vaguely worded endorsement could cover less than it appears to ^[1]. At each renewal, ask your broker whether any AI-related endorsement, exclusion, or definition has been added, and get an explanation of what it changes. Firms that track their policy language as carefully as they track their AI usage will not be surprised by either.

Reading policy wording, not headlines

The gap between what a headline says and what a policy does is where a lot of avoidable anxiety lives. A headline announcing that AI exclusions are spreading is accurate in the sense that such exclusions are appearing in some lines, and misleading in the sense that it invites a lawyer to assume their own malpractice coverage has changed when the reporting available does not show that for standard LPL forms. The only reliable cure for that gap is to read the actual wording of the actual policy, including any endorsements, rather than to reason from the general climate.

Broad language deserves particular attention because it is ambiguous by design or by haste, and ambiguity is resolved at the worst possible time, which is after a claim. Pillsbury's caution that early AI exclusions are broadly worded with uncertain impact is a reminder that the words on the page, not the intent behind them, are what a firm will ultimately rely on ^[1]. When an endorsement uses a term like artificial intelligence without defining it, the firm and the carrier may hold very different pictures of what it covers, and that difference becomes a dispute rather than a footnote once money is at stake.

The habit worth building is simple. Treat every renewal as an occasion to read what changed, ask your broker to explain any AI-related language in plain terms, and record the explanation. A firm that maintains that habit will always know where it stands, which is a calmer and safer place to be than reacting to whatever the market happens to be saying that quarter.

Where RankShield Legal fits, stated honestly

It would be easy, and dishonest, to present verifiable AI documentation as a way to buy down a premium. It is not, and no vendor should claim otherwise. Pricing and coverage decisions belong to carriers, and they weigh many factors that have nothing to do with any single control. What documentation does is more modest and more durable: it lets a firm answer the specific questions carriers are already asking with records instead of adjectives, and it gives the firm the same records to rely on if a matter is ever contested.

RankShield Legal is a vendor, not a law firm, and this article is informational rather than legal or insurance advice. Within that boundary, the platform's contribution is narrow and concrete. It produces signed records of certified citation checks performed before filings, and attested records that privileged material never reached third-party models. Those are exactly the two areas where the renewal questionnaire presses hardest, and they are areas where a firm otherwise has only its own account to offer. Turning that account into a record is the whole of what the tool claims to do.

The honest summary is that governance evidence does not guarantee coverage, does not lower a quote, and does not replace the judgment of the attorneys who remain responsible for their work. It supports the conversation a firm has with its insurer, and the conversation it may later have with a court, by grounding both in verifiable facts. For a firm that wants to answer AI questions from a position of evidence rather than assurance, that support is worth having, and it is worth understanding for what it is rather than for what a louder pitch might claim.