The question to settle before your next renewal
If ransomware or wiper malware linked to a foreign government shuts down your operations, will your cyber policy pay? The answer sits in one endorsement most buyers skim past: the cyber war exclusion. Since Lloyd's tightened its requirements in 2023, that clause has carried more weight than almost any other wording in a standalone cyber policy. Two policies with identical limits and premiums can respond very differently to the same state-linked incident.
You probably aren't the target. The realistic scenario: malware aimed at another country's infrastructure spreads through a shared software vendor, a managed service provider or an unpatched VPN, and your plant, clinic or logistics hub goes dark as a bystander. How your wording treats that bystander loss matters far more than how it treats a declared war.
This piece covers how the Lloyd's market model clauses work, how carrier-specific wordings differ, what attribution and "impacted state" language actually do, and which carve-backs are worth your negotiating capital.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Why the old war exclusions stopped working
Traditional war exclusions were written for physical conflict: invasion, hostilities, "warlike action." They never anticipated destructive code released by an intelligence service in peacetime and spreading globally within hours.
The 2017 NotPetya outbreak exposed that gap. Two large disputes followed: Mondelez against Zurich, and Merck against its property insurers. Both turned on whether a hostile or warlike action exclusion applied to malware widely attributed to a state. Both were eventually resolved by settlement, and in the Merck case a New Jersey court had held that the traditional exclusion didn't reach the attack. Insurers drew a clear lesson: the old wording was too ambiguous to rely on, and one event could trigger thousands of claims at once without being priced for it.
Lloyd's responded with a market bulletin requiring standalone cyber policies written by its syndicates to carry clear exclusions for state-backed cyber attacks, effective in 2023. The wording must exclude war losses and state-backed attacks that significantly impair a state's ability to function, and it must say how attribution is determined. Many carriers outside Lloyd's, across Europe, Asia, the Middle East and the Americas, adopted their own versions, so the issue reaches well beyond London.
How the LMA cyber war exclusion clauses differ
The Lloyd's Market Association published four model clauses, LMA5564 to LMA5567, and has issued revisions since. They run roughly from most restrictive to most generous. The table is simplified; read the exact clause and any amendments in your own policy.
| Clause | What it excludes | What survives | Buyer takeaway |
|---|---|---|---|
| LMA5564 | War and all cyber operations | Very little for state-backed events | Push back hard. This is the weakest position for a buyer. |
| LMA5565 | War and cyber operations, with a sublimit option | Cyber operation losses up to a stated sublimit, with retaliatory operations between major powers excluded | Acceptable only if the sublimit is meaningful relative to your BI exposure |
| LMA5566 | War, retaliatory operations between specified states, and operations with a major detrimental impact on a state's essential services or security | Other cyber operations at full limit | A reasonable middle ground for many mid-size firms |
| LMA5567 | As 5566, with narrower application | Protects bystanding cyber assets and losses outside the impacted state | The clause to ask for, especially if you operate across several countries |
Carrier-specific wordings don't map neatly onto this ladder. Some large insurers use their own endorsements that tie coverage to the scale of an event. A "widespread event" might carry a lower sublimit, a higher retention or coinsurance, while a contained state-linked attack on your company alone is paid in full. Those wordings aren't automatically worse than LMA5567, but they shift the key question from "who did it" to "how many others were hit." Your broker should model both.
Attribution and the impacted state, in plain terms
Two concepts decide most disputes.
Attribution. The LMA clauses look first to whether the government of the state where the affected system is physically located attributes the operation to another state. Until that happens, the insurer may rely on an objectively reasonable inference from available evidence. Under the model wordings, the insurer carries the burden of proving the exclusion applies. That matters for any attribution cyber insurance claim: with servers in Singapore, the UK and Brazil, three governments may say different things, or nothing, and formal attribution can take months or never arrive.
Impacted state. This is a state where a cyber operation has had a major detrimental impact on its functioning, through disruption of essential services such as financial infrastructure, health, utilities or security and defense. The more generous clauses exclude losses tied to that impact but protect systems outside the impacted state and bystanding assets that weren't the target.
A hypothetical worked example. Say a 400-person manufacturer has plants in Poland and Mexico. A state-linked wiper hits Polish energy infrastructure and spreads through a shared industrial software update, encrypting systems at both plants. Under LMA5564, both losses are likely excluded. Under LMA5567, the Mexican plant's interruption is likely covered as a loss outside the impacted state. The Polish plant is the contested one, depending on whether Poland qualifies as an impacted state and whether the plant's systems count as bystanding assets.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Collateral damage is where your exposure really sits
For most companies, state-backed cyber attack insurance isn't about being singled out by an intelligence agency. It's about spillover. State-linked groups also run financially motivated ransomware, and the line between criminal and state activity is often blurry. An insurer facing a large aggregated loss can point to exactly that blurriness.
The exposures most likely to bite:
- Business interruption from wiper or ransomware spreading through common software, cloud dependencies or managed service providers
- Contingent BI when a supplier or IT provider in another country is hit
- Data restoration costs, which can run from tens of thousands to several million USD for mid-size firms depending on backups and architecture
- Incident response fees incurred before anyone knows who's behind the attack
Companies with operations, suppliers or cloud regions near active geopolitical conflict, such as Eastern Europe, the Gulf, East Asia or South Asia, carry higher bystander risk. They should weight this clause more heavily than a firm operating in a single low-tension market.
Carve-backs worth negotiating
Leverage depends on your size, loss history and controls, but these asks are realistic in a competitive market:
- Move to LMA5567 or an equivalent. If the insurer offers 5564 or 5565, ask what stepping up would cost. The premium difference is often modest, though it varies by contract and region.
- Bystander and outside-impacted-state protection, stated explicitly, including contingent BI.
- Incident response and defense costs paid until attribution is established, with no clawback unless the exclusion is finally proven to apply.
- A clear attribution standard that relies on government attribution, not the insurer's own forensic view alone.
- A meaningful cyber operation sublimit if a full carve-back isn't available. As a rough rule of thumb, a sublimit well below half your main limit deserves scrutiny.
- Narrow definitions of "essential services" and "major detrimental impact", so a sector-specific incident doesn't automatically become a state-level one.
Red flags include a war exclusion with no attribution mechanism, an insurer-determined attribution standard, a "widespread event" trigger with no numerical threshold, and a war exclusion in a property or liability policy that conflicts with the cyber wording.
Wording interpretation is a legal question. Ask a specialist broker and, for larger programs, coverage counsel licensed in your policy's governing jurisdiction to review the final endorsement.
A decision rule for your next renewal
Ask your broker for three things at least 60 days before renewal: the exact war and cyber operation exclusion wording proposed, a side-by-side comparison against LMA5567 or the most generous equivalent available, and a written explanation of how attribution would work in each country where you hold systems.
Then apply a simple rule. If a large share of your revenue, say a third or more, depends on systems or suppliers outside your home country, treat bystander and outside-impacted-state protection as a must-have and pay for it if needed. If you're a single-country business with limited cloud concentration, a solid LMA5566-style wording with a clear attribution standard is usually adequate, and your effort is better spent on BI waiting periods and sublimits.