Does your malpractice insurance cover AI mistakes? What the 2026 exclusions actually hit

Yes, almost certainly. As of September 2026 a US lawyers professional liability policy still responds to a malpractice claim caused by an AI error, and in England and Wales an SRA-qualifying policy is not permitted to exclude one at all. The generative AI exclusions filling your inbox attach to commercial general liability, directors and officers, and errors and omissions forms. They are not on your malpractice paper. The gap worth your attention is older and duller than an exclusion: your policy has never paid court sanctions, and a sanction is precisely what a fabricated citation produces.

The exclusions everyone is forwarding sit on a different policy

Verisk’s ISO arm put a set of generative AI exclusions into circulation with a January 2026 edition date. The broad one is CG 40 47, which knocks out bodily injury, property damage and personal and advertising injury arising out of generative AI under the Commercial General Liability Coverage Part. CG 40 48 is the narrower sibling, taking only Coverage B. A third form, CG 35 08, does the same job on the Products and Completed Operations Coverage Part, which has nothing to do with a law firm. All three carry the same definition: a machine-based learning system or model trained on data with the ability to create content or responses.

Look at where those forms land. Coverage A of a CGL policy is bodily injury and property damage. Coverage B is personal and advertising injury. A client suing you over a badly drafted indemnity clause is claiming economic loss from professional services, which is neither. Your general liability policy was never going to pay that claim, with or without an AI endorsement stapled to it.

The other exclusion in circulation is Berkley’s, reported by Hunton’s insurance recovery team as the first “absolute” AI exclusion, barring any claim based upon, arising out of, or attributable to any actual or alleged use, deployment, or development of artificial intelligence. It is genuinely aggressive. It is also, per the same report, intended for the company’s D&O, E&O and fiduciary liability products. When your broker forwards a warning about AI exclusions, the first question is not how broad it is. It is which policy it attaches to.

What the malpractice market is actually doing

On 25 August 2026 the Maryland Daily Record put the question directly to people who underwrite and place lawyers cover. The answers are more useful than any tracker page.

Al Roberts, who runs professional liability for Berkley Select, said his company has no AI exclusionary language in place and is not currently asking law firms supplemental underwriting questions about AI use. Hold that next to the paragraph above. The corporate family whose absolute AI exclusion is being cited everywhere as proof that cover is vanishing has an LPL arm that has not written one. That single fact should recalibrate most of what you have read on this topic.

Lee Norcross of L Squared, a broker that specialises in lawyers professional liability nationally, said he is aware of exactly one nonstandard surplus lines insurer carrying an AI exclusion endorsement, and called it the exception rather than the rule. Chris Newbold, COO of ALPS, said the carrier is not seeing direct AI claims yet, only matters where AI turns up as part of the fact pattern behind a citation problem. He described that as an early signal rather than a clear trend.

Two things follow. You have a window, and it is not going to stay open. Newbold expects ALPS to start asking detailed questions about firm AI use, slowed only by the fact that applications and forms need state regulatory approval. What underwriters said they will price is not whether you use AI. It is how you govern it, which is the argument for treating your training record as an insurance asset rather than a compliance chore.

The loss you will actually suffer is the one nobody insures

Here is the part the coverage-panic articles skip. A malpractice policy pays to defend covered claims and to indemnify covered losses. It typically does not pay court-imposed sanctions, disciplinary penalties or fines.

Now look at what has actually gone wrong for lawyers using AI since 2023. Not a wave of clients suing over defective AI-drafted advice. A steady run of judges imposing sanctions for invented authority, plus referrals to disciplinary bodies. The exposure that has materialised is sitting squarely in the one category your policy was always going to decline. As Norcross put it, AI does not have a law license to lose. Neither does it have a policy limit.

There is one narrow place where the CGL exclusions do reach a small firm, and it is worth checking. Coverage B is where a claim for copyright infringement or defamation arising from your own published material would sit. If your firm publishes marketing copy or images with AI in the loop, CG 40 48 removes precisely that cover. It is a small exposure compared with a negligence claim. It is also the only AI exclusion in the current wave that has any bearing on you.

In England and Wales your insurer is not allowed to exclude AI

Solicitors get a structural protection that US lawyers do not, and almost nobody writing about AI insurance mentions it. Clause 6 of the SRA Minimum Terms and Conditions opens by saying the insurance must not exclude or limit the insurer’s liability except to the extent a claim arises from the matters set out in that clause. It is a closed list. Prior cover, bodily injury, property damage, partnership disputes, employment claims, trading debts, fines, dishonesty, directors’ liability, war and asbestos, sanctions, and cyber. Artificial intelligence does not appear anywhere on it.

The cyber permission at clause 6.12 is the closest an insurer could get, and it comes with a proviso that guts the attempt: any such exclusion must not limit the insurer’s liability for civil liability arising from breach of duty in the performance of legal work. Clause 4.11 finishes the job by requiring that any provision inconsistent with the Minimum Terms be severed or rectified. An AI exclusion endorsed onto a qualifying policy is not a negotiation. It is void.

Two limits on that comfort. First, the qualifying layer is £3 million for a recognised or licensed body and £2 million otherwise. Excess layers above it are ordinary commercial insurance and can carry whatever wording the market will write, so ask your broker for the excess layer terms specifically rather than the summary. Second, clause 6.7 excludes fines and penalties, which means a solicitor sanctioned over an AI-fabricated citation is in exactly the same uninsured position as a lawyer in Ohio.

What to do before your next renewal

Five things, in the order I would do them.

• Pull your LPL policy and search the wording for “artificial intelligence” and “machine learning”. If neither appears, you have no AI exclusion and you can stop worrying about the tracker pages. If it does appear, get the endorsement number and send it to your broker with a request to quote it away.
• Check Coverage B on your general liability policy for a CG 40 48 or CG 40 47 endorsement, but only if your firm publishes marketing content with AI in the loop.
• Answer the renewal application AI question in writing, describe what the firm actually does, and keep the copy. Insurers are already using AI to verify what applicants submit.
• Have a written AI policy and a dated training log before the question arrives, because governance is the variable underwriters said they will price. Our policy template is a starting point, not a substitute for the training.
• Assume sanctions come out of your own pocket. Price your verification time accordingly.

My view, stated plainly: stop shopping for AI insurance and start budgeting for verification. The insurance question has a boring answer in both jurisdictions right now, and the boring answer is that you are covered for the claim and exposed on the sanction. The firms that get hurt in the next eighteen months will not be the ones whose policy had an exclusion. They will be the ones who filed something they did not read.

If you want the underlying controls in one place, our free guide covers the verification and confidentiality steps that keep a claim from happening at all, and our tool guide covers which products run on terms a small firm can actually defend to a regulator.

FAQ