Chubb Is Excluding the Risk Its Own CEO Says AI Will Solve
Six carriers filed AI liability exclusions. Two Lloyd's MGAs are underwriting what they dropped
Carriers are actively excluding AI liabilities from their standard products. Since January 1, 2026, Verisk/ISO exclusion endorsements have allowed carriers to remove generative AI exposure from commercial general liability policies, and at least six major carriers have filed their own AI exclusions with state regulators. GenAI lawsuits in the U.S. grew 137% year over year in 2024 to 2025, with cumulative filings past 700. The liability is real, it is accelerating, and it is flowing to the enterprise deployer because every major AI vendor caps its own exposure at 12 months of fees. The same MGA model that Chubb’s CEO publicly argued AI will replace is now the model underwriting the AI liability class that Chubb’s standard products cannot cover.
Today’s Deep Dive covers:
What Does the AI Liability Exposure Actually Look Like, and Why Can’t Existing Carrier Products Cover It?
Why Is the MGA Model Structurally Better Positioned for AI Underwriting Than the Carrier Direct Model?
What Has to Be True for AI Liability MGAs to Get Capacity, and What Breaks If They Can’t?
1. What Does the AI Liability Exposure Actually Look Like, and Why Can’t Existing Carrier Products Cover It?
A new liability class is forming in the gap between what enterprises are deploying and what their insurance programs actually cover, and as of January 1, 2026, that gap became structural.
On that date, Verisk/ISO exclusion endorsements took effect allowing carriers to remove generative AI liabilities from commercial general liability policies. Form CG 40 47 excludes bodily injury, property damage, and personal and advertising injury arising from generative AI across both occurrence and claims-made CGL coverage. A parallel form, CG 35 08, applies the same exclusion to products and completed operations. Verisk’s ISO form templates serve as the standard policy language foundation across the U.S. commercial liability market, and the company reported strong carrier interest even before the effective date. WR Berkley went further with Form PC 51380, an absolute AI exclusion for D&O, E&O, and fiduciary liability that eliminates coverage for any claim arising from the use, deployment, or development of artificial intelligence by any person or entity. AIG, Great American, Hamilton Insurance Group, and Philadelphia Indemnity have all filed their own AI exclusions with state regulators. This extends the exclusion trend we first covered in our January 27 digest and represents exactly the structural bifurcation that coverage was tracking toward: conventional insurers withdrawing from silent AI exposure while a small cluster of specialist underwriters moves into the gap.
The exclusions are a response to exposure that carriers cannot price. Gallagher Re’s March 22 white paper with MIT and Testudo found that AI liabilities sit in the gaps between every major existing product line. Cyber insurance may cover an AI-enabled phishing attack but does not cover liabilities from AI outputs such as hallucinations, defamation, or intellectual property infringement. Technology E&O protects technology providers, not deployers, leaving most enterprises using third-party AI tools structurally uncovered. CGL was never designed for algorithmic failures. The result, as the Gallagher Re report states directly, is that AI introduces a new class of risks that traditional insurance policies were never designed to address. Lockton Re’s February 2026 report, produced with Armilla AI, reached the same conclusion independently: CGL insurers do not currently model, underwrite, or price AI risks, and there is a growing gap between what insurers intend to cover and what they actually cover based on the policy language.
The litigation trajectory confirms the exposure is not theoretical. Testudo’s proprietary database, cited in the Gallagher Re paper, tracks cumulative U.S. GenAI lawsuits past 700 between 2020 and 2025, with filings accelerating 137% year over year in 2024 to 2025. The leading claim categories: patent infringement at 11.9% of cases, copyright infringement at 11.2%, and personal injury tied to privacy violations at 10.2%. Anthropic settled copyright claims in August 2025 at approximately $3,000 per infringing work, establishing a damages benchmark for future litigation. Aon’s Kevin Kalinich told the Financial Times that the industry can absorb a single company’s AI failure but cannot absorb a systemic event: a flaw in a widely deployed foundation model triggering thousands of correlated claims simultaneously.
The structural driver making this an insurance problem rather than a legal problem is vendor contract design. All four major AI vendors, OpenAI, Anthropic, Google, and Microsoft, cap their liability at 12 months of fees paid and disclaim all consequential damages. No vendor offers performance warranties under standard terms. The deployer, the enterprise customer that is the existing commercial insured, bears the residual liability for everything the AI system produces. As Gallagher Re concluded: AI vendor contracts favor the vendor and leave deployers shouldering most of the liability.
This follows TIC’s March 25 deep dive on Chubb’s automation roadmap and the question of whether AI replaces or redefines the MGA model. The answer is becoming clearer: AI is simultaneously the tool carriers are deploying to automate underwriting and the emerging liability class that their standard products cannot cover.
