Insurers rewrite AI wording

The fight over AI in insurance has moved out of the lab and into the policy jacket. Commercial carriers and brokers are no longer asking only whether generative AI can write claims notes, summarize medical records, or help underwriters sort submissions. They are rewriting the sentences that decide whether any of the damage from those tools will be covered at all. Business Insurance reported on April 7 that insurers and brokers are updating contract language and exclusions as AI use spreads through commercial accounts, with the first signs of affirmative AI coverage appearing alongside new carve-outs that narrow it (businessinsurance.com). That shift sounds dry until you picture the moment a claim arrives. A company uses an AI tool to generate marketing copy, and a rival sues for defamation or copyright infringement. A claims team relies on an AI summary that misses a key fact. A broker helps place cyber or liability coverage for a client that has AI stitched into customer service, fraud screening, and pricing. In each case, the fight may turn less on what the model did than on whether the policy’s words treat that act as ordinary software risk, excluded cyber risk, or a new AI-specific exposure (businessinsurance.com; hunton.com). The wording is changing because the old forms were written for a different machine age. Verisk said in a July 2025 filing that its ISO general liability update would add optional endorsements to help insurers address “developments related to generative AI,” with a proposed effective date of January 1, 2026 (core.verisk.com). Independent Agent, summarizing the filing for agents, said the new endorsements let carriers exclude generative AI exposures under general liability policies, which is a blunt way of saying that losses many buyers assumed would fall under familiar coverage may now be pushed outside the fence (independentagent.com). Once those forms exist, the market starts to sort itself. Some insurers adopt the exclusions. Some brokers try to negotiate them back. Some specialty players build policies that cover the gap for companies that make or deploy AI systems. That is why this moment looks less like a verdict on AI itself than a repricing exercise. Insurance has begun treating AI the way it treats any fast-growing source of loss: first by tightening definitions, then by selling back selected pieces of protection at a price (businessinsurance.com; lexology.com). Inside carriers, the language work is happening alongside a second push: keep a human close to the decision. Digital Insurance reported on April 8 that insurance technology executives using AI in underwriting and risk assessment are emphasizing governance, accountability, and human judgment for more complex risks rather than letting models run unattended (dig-in.com). The pattern is visible in regulation too. The NAIC’s model bulletin tells insurers that decisions made or supported by AI still have to comply with existing insurance law, including unfair discrimination rules (content.naic.org). New York’s Department of Financial Services went further in July 2024, telling insurers that if they use AI or outside data in underwriting and pricing, they need governance, testing, and oversight strong enough to show the systems are not producing unlawful discrimination (dfs.ny.gov). That combination — narrower coverage on the outside, tighter controls on the inside — explains the mood of the market. Insurers still want more AI. They just want it fenced, logged, reviewed, and named correctly in the contract. Even some of the industry’s more futuristic talk now comes wrapped in that caution. Reinsurance Group of America told Insurance Business that more advanced AI could eventually replace today’s patchwork of separate tools across underwriting, actuarial work, and compliance, but that vision sits far ahead of the immediate work of governance and implementation (insurancebusinessmag.com). For brokers, underwriters, and claims leaders, this turns AI from a tech project into a reading project. They have to know where a client uses AI, which vendors sit underneath it, what a form now excludes, and which decisions still require a person to slow down and look again. The industry’s newest AI tool may be a red pen dragged through a liability endorsement, one sentence at a time (businessinsurance.com; dig-in.com).