Executives who think they understand their cyber war exclusion are usually working from a case that no longer describes their policy. The case is Merck's. The policy language has moved on without most boardrooms noticing.
This article is grounded in current advisory work, not retrospective analysis. Mark Lynd is a 5x CEO/CIO/CISO with Thinkers360 Top 10 global rankings across Cybersecurity and Artificial Intelligence and was ranked #1 globally in Cybersecurity in 2023. He is currently Head of Executive Advisory and Strategy at Netsync, advising enterprise C-Suites and boards on the AI and cybersecurity questions moving fastest in 2026. The frameworks and patterns referenced here are from active engagements this quarter.
Misconception One. The War Exclusion Rarely Applies, So It Rarely Matters.
The 2017 NotPetya attack hit Merck hard, damaging more than 40,000 of the company's computers and disrupting manufacturing, sales, and research operations. Merck filed a $1.4 billion insurance claim. Its property insurers denied roughly $700 million of that claim, pointing to a war exclusion clause and to the U.S. and U.K. governments' public attribution of NotPetya to Russian military intelligence. In May 2023, a New Jersey appellate court sided with Merck, ruling that a war exclusion written for "hostile or warlike action" requires actual military action, not merely a cyberattack carried out by a government-linked actor with hostile intent. Merck and its insurers settled in January 2024, just before oral arguments on further appeal, on terms that were not disclosed.
Executives who followed that story often concluded the war exclusion is close to toothless for cyber claims. That conclusion was correct the moment the ruling landed, for a policy written the way Merck's was. It is the wrong conclusion for a policy bound today, and that gap between the headline and the current contract is the first thing executives get wrong.
Misconception Two. Winning That Argument Once Means It Stays Won.
The market did not hold still after the Merck ruling. Lloyd's of London required its managing agents, effective March 31, 2023, to write explicit state-backed cyberattack exclusions into every standalone cyber policy, using one of four model clauses published by the Lloyd's Market Association. These clauses do not rely on the ambiguous "warlike action" phrasing that helped Merck win its case. They exclude losses from cyberattacks attributed to a state where the attack materially impairs a state's ability to function or its security capabilities, and the later version of the clause hands the attribution decision to the insurer directly rather than requiring outside proof.
The four model clauses vary widely in how much room they leave policyholders. One is comparatively narrow, protecting losses to systems physically located outside any state directly targeted. Another excludes coverage with no carve-back at all. An executive who cites Merck as evidence that the war exclusion cannot be used against them is reading a court's interpretation of vague language written years before this newer, deliberately unambiguous language existed. Merck won an argument about how to read an old sentence. The market's answer was to stop writing that sentence.
Misconception Three. The Real Exposure Is the Rare Nation-State Event.
Boards spend a disproportionate amount of governance time worrying about the catastrophic, headline scenario, a nation-state actor knocking out core systems the way NotPetya did. That scenario deserves attention, but it is not where most companies are actually losing coverage today. The Insurance Services Office introduced a generative AI exclusion into commercial general liability policies in January 2026, and cyber and management liability carriers are following with their own AI-specific endorsements, new application questions, and narrower base coverage forms, according to legal analysis published by Fenwick. These changes rarely arrive as one clause anyone would flag in a five-minute read. They show up as a policy definition that quietly moves AI-related losses outside the base cyber form and into a separate rider most companies never purchased.
For the overwhelming majority of policyholders, an everyday AI-related incident, a vendor's copilot leaking customer data, a chatbot's hallucinated output creating a liability claim, is a far more probable loss than a nation-state cyberattack triggering a war exclusion fight. Treating the rare scenario as the main event while an ordinary one quietly falls outside coverage is the third and most common misconception, because it is the one almost nobody is checking for at renewal.
Part of why this misconception persists is that war exclusion disputes generate headlines and AI exclusion endorsements do not. A three-year court fight over a $700 million claim gets covered by every trade publication in the industry. A new definition added on page eleven of a renewal endorsement gets covered by nobody, including, in most cases, the company's own legal team, who reviews the base policy closely and treats endorsements as boilerplate. Underwriters know this. It is one reason coverage erosion increasingly happens through endorsements and application questions rather than through a clause dramatic enough to make the news.
A Worked Example
A mid-market financial services firm with roughly $300 million in revenue spent two full board meetings in 2025 debating what would happen if a nation-state actor hit its systems, and directed the CISO to negotiate the war exclusion language specifically. The broker delivered a narrower, more favorable war clause. Meanwhile, the same 2026 renewal quietly added a new AI systems exclusion endorsement, moving any loss tied to an AI tool outside the base cyber policy. Nobody on the renewal committee read that endorsement closely, because the board's attention that cycle was on war language.
Eight months later, a vendor's AI-powered support copilot leaked a batch of customer records through a misconfigured training pipeline. The claim was initially denied under the new AI exclusion. The broker eventually negotiated a partial payment, but only after months of dispute and legal cost the company had not budgeted for. The nation-state scenario the board spent its attention on never happened. The everyday one nobody was watching did.
The Counterargument Worth Taking Seriously
Someone could reasonably argue this framing undersells real war exclusion risk. Nation-state cyber activity against critical infrastructure, defense contractors, and firms with a clear geopolitical target profile is rising, and for those organizations the war exclusion is not a hypothetical, it is the single most consequential clause in the policy. A defense subcontractor or a utility operator that treats war exclusion negotiation as a secondary concern is making a real mistake, and that argument deserves full weight rather than a dismissal.
The distinction is target profile. For a company that state-linked actors have genuine reason to target, war exclusion language earns board-level attention and a broker who specializes in exactly that clause. For the median company outside that profile, spending the same governance effort on a scenario claims data shows is rare, while an AI exclusion quietly narrows coverage for a loss type that is common, is a misallocation the data does not support.
A practical way to resolve this without picking a side in the abstract is to have the broker map both exposures side by side, the specific war exclusion clause bound in the current policy and the specific AI-related carve-outs added at the last renewal, and bring both to the same board conversation. Most companies currently review these on entirely separate tracks, if the AI language gets reviewed at all, which is exactly how the second exposure ends up invisible.
Questions for Monday
Ask which Lloyd's Market Association model clause, if any, sits in the current policy, and what it actually excludes. Ask who inside the company has read the AI exclusion or endorsement added at the last renewal, in full. Ask whether the incident response retainer specifies who determines nation-state attribution and on what timeline, since that determination can decide whether a claim gets paid. And ask honestly whether leadership and the board are spending governance time on the rare catastrophic scenario or on the everyday exclusion the claims history says is more likely to matter.
Merck proved a court could read an old sentence generously. It did not buy the rest of the market the same sentence.
