EPIC Insurance Brokers' 16th Annual Lawyers' Professional Liability Claims Survey, released in May 2026, reports that seven of 13 surveyed carriers saw an increase in AI-related claims over the past year. Those 13 carriers collectively insure more than 80% of Am Law 200 firms.
It is the first credible loss emergence data for AI in any professional liability line. It arrives while the coverage forms are mid-revision and the exposure has no development history to price against.
Key Takeaways
- Seven of 13 carriers (54%) reported AI-related claims increases, the first loss emergence data specific to AI in professional liability.
- Eight of 13 (62%) reported higher overall claim frequency, the first rise in five years for the LPL market, and eleven of 13 (85%) reported materially higher defense spending.
- 1,227 documented court cases of AI-fabricated legal work product worldwide, 811 in the United States, growing at five to six new cases a day.
- 89.9% of documented incidents involve firms under 25 attorneys, with solo practitioners alone accounting for 50.4%.
- No major U.S. LPL writer has filed an explicit AI exclusion on a named lawyers' professional liability form as of May 2026, while exclusions proliferate in management liability and general liability.
What the EPIC Survey Found
The survey's three headline splits move together. Seven of 13 (54%) report AI-related claims rising. Eight of 13 (62%) report higher overall frequency, the first increase in five years. Eleven of 13 (85%) report materially higher defense spending.
AI claims are expensive to defend for structural reasons rather than incidental ones: they raise novel questions of professional competence, require technology-specific discovery, and typically run three proceedings in parallel, the malpractice suit, the underlying case, and a bar disciplinary action.
The severity backdrop is already heavy. Nine-figure claims are "no longer viewed as rare outliers" per the survey, and EPIC's historical work across $5.2 billion of analyzed gross written premium from 2006 to 2018 shows $7.8 billion of severe losses at the $5 million-plus threshold, with accident year 2018 alone producing one claim above $100 million and multiple above $10 million.
Eileen Garczynski, who leads EPIC's Law Firm Group, put the underlying principle in one line: "The duty of competence cannot be delegated to technology."
The line has margin to absorb this, which is part of the problem. AM Best's LPL composite of 16 specialty insurers wrote $728 million in direct premium in 2025, up from $709 million, with cumulative growth above 18% since 2020 after near-flat growth below 1% from 2015 to 2019. The composite's 2024 operating ratio was 58.7 against 84.7 for commercial casualty.
Pricing an Exposure With No Triangle
The claims pipeline is visible in a way the losses are not. Damien Charlotin's Hallucination Database at the HEC Paris Smart Law Hub documents 1,227 cases globally, 811 of them in the United States, up from roughly 200 a year ago and 719 in January 2026. That is five to six new documented cases a day.
| Case | Court | Date | Details | Sanctions |
|---|---|---|---|---|
| Mata v. Avianca, Inc. | S.D.N.Y. | June 2023 | Six fabricated cases from ChatGPT | $5,000 fine |
| Park v. Kim | Second Circuit | 2024 | AI-generated fake citations | Referral for discipline |
| Kruse v. Karlan | Missouri Court of Appeals | 2024 | Multiple AI-fabricated citations | Appeal dismissed |
| Whiting v. City of Athens | Sixth Circuit | March 2026 | 24+ fake citations, defied show cause order | $15,000 per attorney + full fee reimbursement + double costs |
| Mississippi dual-side case | Federal court, Mississippi | November 2025 | Both sides independently submitted AI-fabricated briefs | $3,500 + 2-year bar (defense); $2,500 + CLE (plaintiff) |
Composition matters more than the count. Of the 1,227 incidents, 1,022 involved fabricated case citations and 323 involved false quotes attributed to real cases. Solo practitioners account for 50.4%, and firms under 25 attorneys for 89.9%. The exposure therefore concentrates in small accounts, where a fixed defense cost consumes a far larger share of the policy premium than the same cost would at an Am Law firm, so severity compounds against the smallest premium base.
Sanctions themselves are modest, the largest single penalty on record standing at $109,700. They are not the loss. Each incident can generate a malpractice suit from the affected client, a disciplinary proceeding, and reputational damage across the firm's book, and on the pattern observed so far those follow-on claims emerge six to eighteen months after the initial sanction.
That lag is what makes this unpriceable by normal methods. The earliest documented case, Mata v. Avianca, dates to June 2023, so even if every sanction had produced an immediate claim there would be at most three accident years of data and no tail factors at all. Chain-ladder and Bornhuetter-Ferguson both require development history that does not exist for this exposure, and the usual fallback of borrowing from an analogous class runs into a moving causal mechanism: a model that fabricates citations in 2024 may not do so in 2026 after architectural changes, or may fail differently.
The market is pricing anyway. AI-specific endorsements run 5% to 15% above base premium, $2,500 to $7,500 for a mid-size firm paying $50,000, against a single AI-related event capable of generating $500,000 to $2 million in defense costs and damages. That is a rate set on judgment overlays, the same approach cyber took between 2005 and 2010, with the same failure mode: if the initial frequency or severity assumption is light, the underpricing compounds across renewal cycles before credible data forces the correction.
The Forms Diverge Faster Than the Data Accumulates
The second problem is that even when development arrives, it may not be usable across books.
As of May 2026, no major U.S. LPL writer has filed an explicit AI exclusion on a named lawyers' professional liability form. CNA added supplemental AI questionnaires at renewal and flagged AI as a novel claims exposure in its fiscal 2025 10-K. AmTrust carries the only confirmed verbatim AI question on a U.S. LPL application. Nine carriers surveyed have no public AI position at all.
Meanwhile exclusions are proliferating one line over. Hamilton Select filed the broadest known professional liability AI exclusion, naming ChatGPT, Bard, Midjourney and DALL-E by product. W.R. Berkley's Form PC 51380 applies an absolute AI exclusion to D&O, EPL and fiduciary. On general liability, ISO's CG 40 47 and companion endorsements took effect January 1, 2026, with more than 80% of state filings approved by April. Beazley has said it has no plans to add AI exclusions to its professional liability forms.
The result is that two similarly situated firms with identical AI-related claims can reach opposite coverage outcomes depending on which carrier wrote them: one affirmed under a governance-contingent endorsement, the other litigated under a silent pre-2023 form. Industry-level LPL loss data aggregates across that heterogeneity, so the frequency and severity it eventually shows will reflect a mix of coverage positions rather than a mix of underlying risk, and applying it to a specific book requires knowing which position that book actually holds.
The 58.7 operating ratio is what lets this continue. A line running that far ahead of commercial casualty can absorb emerging AI losses without rate action, which is why the questionnaires arrived before the endorsements and why the regulatory attention landing on AI claims handling is running ahead of the pricing response. Margin buys time to gather data. It also buys time to accumulate a book written at rates that never contemplated the exposure.
Further Reading on actuary.info
- ISO CG 40 47 AI Exclusion Forces a GL Rate Adequacy Rethink - The parallel exclusion trend on general liability lines, including the loss elimination ratio framework for isolating silent AI exposure from historical rate indications after carriers adopt Verisk’s generative AI exclusion endorsements.
- How Actuaries Price AI Liability Coverage When the Loss Triangle Has No Rows - The full actuarial methodology problem the EPIC survey’s 54% figure points to: how carriers build a pricing prior for standalone AI liability coverage when claim counts sit orders of magnitude below the 1,082-claim full-credibility standard.
- Malpractice Insurers Hit 105% Combined Ratio as Verdict Severity Doubles - The medical professional liability market provides a precedent for how verdict severity acceleration and reserve deterioration spread across professional liability lines absorbing new risk categories.
- NAIC AI Claims Handling Regulatory Focus in 2026 - How state regulators are approaching AI-related claims across multiple lines, including the supervisory frameworks that will shape how carriers deploy and price AI endorsements.
- Litigation Funding Disclosure Rules Reshape P&C Claims Economics - Third-party litigation funding dynamics that will accelerate the spread of AI-related professional liability claims as funders identify asymmetric information advantages in AI malpractice cases.
- P&C Claims Severity Faces a Four-Factor Compounding Problem - The broader severity environment in which AI-related professional liability claims are developing, with analysis of how multiple independent severity drivers interact to produce compounding trends.