Insurers are tightening cyber policies as AI‑driven threats emerge. Australian carrier QBE filed wording that caps payouts for “LLMjacking” – attacks that hijack large language models to dodge usage fees. On a $5 million policy, the sublimit would reduce an AI‑related loss to roughly $250,000, about ten percent of the overall limit.

British cyber‑specialist Beazley is drafting similar clauses to limit exposure to regulatory fines tied to AI misuse. Brokers and law firms warn that such sublimits, typically 10 % of the total policy, could shrink coverage for a broader set of emerging AI risks, from model hallucinations to credential theft. Insurers argue the language clarifies what is and isn’t covered.

Market analysts see the move as insurers trying to preserve profitability while AI threatens to inflate claim sizes. AIG and other U.S. carriers have already sought regulator approval to carve AI losses out of existing corporate policies. As firms grapple with the cost of protecting costly AI workloads, the industry may eventually spin off dedicated AI cyber products.

For corporate buyers, the new sublimits translate into tighter risk budgets and potential gaps in protection, prompting many to negotiate bespoke endorsements or retain separate AI‑risk insurance. Brokers like Marsh caution that the trend could drive up premiums across the cyber market as insurers rebalance their loss portfolios. The shift underscores the growing friction between rapid AI adoption and traditional underwriting models.