What cyber insurance doesn't cover in the US: gaps to check
· Updated
Joaquín Trapero in AI and privacy expert, and founder of Nonimo.
A US cyber policy usually leaves out war, your own fraud, bodily injury, physical damage, utility outages, problems you knew about before buying it, and often the money a fraudster tricks you into wiring. That list is everywhere. It settles very little, because claims are won and lost on definitions.
This guide works from two published specimen policies, the market report from state insurance regulators, and the court opinions where a business lost on wording. If a renewal packet sent you here, the AI questions on that form have their own guide.
Cyber insurance typically excludes war, intentional wrongdoing, prior known problems, bodily injury, property damage and infrastructure outages, and it caps wire fraud at a separate, often small, sublimit. Wording differs by carrier and edition, so your declarations page and endorsements are the only answer that counts.
Common cyber insurance exclusions, compared in two real policy forms
Travelers publishes its CyberRisk coverage form CYB-16001 and Hiscox its cyber coverage part CYBCL-CYB P0001A, both 2019 editions marked Specimen. Neither is necessarily the wording you will be sold today. Side by side, though, they show how much a heading like “Infrastructure” can hide.
Filter by what worries you. The columns are my summaries of each clause, not the carriers’ wording, and each row lists the exception that does the work.
Same risk, two specimen cyber forms
| Risk | Travelers CYB-16001 | Hiscox CYBCL-CYB P0001A |
|---|---|---|
| War | Excluded, including undeclared war and military action. A computer attack is carved back unless it supports a war or insurrection | No clause using the word war turned up in this specimen's text; an endorsement can add one |
| Internet, telecom or utility outage | Excluded unless the infrastructure is under your control | Excluded, except a breach of your data held in the cloud or at a hosting provider |
| Laptops or hardware ruined | Tangible property excluded, with a narrow exception under computer fraud | Excluded, except bricking costs, which sit in a list of enhancements and sublimits |
| Upgrading systems after an attack | Paid only under a separate Betterment agreement, with consent and possibly a share you pay | Upgrades excluded, except extra expense, bricking and data recovery |
| Money wired to a fraudster | Social engineering and computer fraud are their own agreements with their own limits. No cover if the sender ever had authorized access to your passwords | Excluded unless the cyber crime coverage applies, and that coverage sits on top of any crime policy you already have |
| Fines and penalties | Not part of covered loss, except regulatory costs and card penalties, where insurable | Regulatory penalties and PCI fines and assessments count as loss, where insurable; other fines do not |
| Fraud or dishonesty by the insured | Excluded, but defense is paid until a final ruling; an executive's conduct counts against the company | Same pattern: defense until final adjudication, and an executive's conduct binds the firm |
| Collecting data unlawfully | Excluded, defense costs only | Counted as a consumer privacy violation; a regulator's case over it falls under the enhanced privacy regulation sublimit |
| Injury claims | Bodily injury excluded; emotional distress kept | Bodily injury excluded; distress from a privacy event kept |
| Patents and trade secrets | Excluded | Excluded, unless caused by a third-party breach or a security failure |
| A problem known before the policy | Liability claims excluded if any executive knew before the knowledge date; acts before the retroactive date excluded | Excluded if you knew before the policy period, measured from your first unbroken Hiscox policy |
From Travelers CYB-16001 Ed. 01-19 and Hiscox CYBCL-CYB P0001A CW (10/19), both specimens, read November 7, 2026. Your policy may use a later edition or add endorsements.
Two things surprised me. Neither form excludes employee mistakes as such, though many online lists say so. And they split on unlawful data collection: Travelers pays only defense, while Hiscox can pay a regulator’s penalties under a sublimit. For Illinois employers under BIPA, that difference deserves a lawyer’s reading.
Does cyber insurance cover wire fraud? Usually through a separate sublimit
The NAIC’s 2026 report says standard cyber policies often exclude or sharply limit social engineering and transfer fraud, leaning on separate crime cover. Coalition reported that business email compromise and funds transfer fraud made up 58 percent of its 2025 claims, worldwide. If the hijacked mailbox also held client files, the breach question comes first.
Each carrier files this loss in a different drawer. Before a renewal call I would want to know which drawer mine uses, and how big it is. Whatever the answer, a callback on any change of bank details, to a number already on file, costs nothing.
| Carrier | Where wire fraud coverage sits | Document |
|---|---|---|
| Chubb | Computer fraud, funds transfer fraud and social engineering fraud, all by endorsement | Chubb cyber products page |
| Hiscox | Cyber crime coverage, excess over any other crime insurance | Specimen CYBCL-CYB P0001A |
| Travelers | Named insuring agreements, plus its own supplemental application, CYB-14302 | Specimen CYB-16001 |
| Coalition | Base form, which took in 11 former endorsements and deepfake-enabled transfer fraud for US surplus lines quotes from April 15, 2025 | Coalition announcement, April 9, 2025 |
A sublimit and a retention can turn a covered loss into a mostly uncovered one. Put your own numbers in; the starting loss is Coalition’s average for transfer fraud that began as email compromise, and the other two are an example.
What a wire fraud sublimit pays
The sublimit, not the headline limit, is what stops this claim.
Sublimit and retention are an example. This assumes the retention comes off the loss first; if your form takes it out of the sublimit, the same example pays $40,000. Loss figure: Coalition 2026 Cyber Claims Report, 2025 data.
Social engineering claim denied: Gore v. Spinnaker (2026)
A Mississippi law firm, Gore, Kilpatrick & Dambrino, bought a policy with a Social Engineering Coverage Endorsement. In May 2024 a man using the name David Casteel hired it to collect a $158,850 debt. A check arrived, the firm wired $158,425 back to him, and the check bounced on June 10, 2024.
The endorsement covered instructions purporting to come from someone who exchanges goods or services with the firm for a fee. On March 31, 2026, the federal court in Mississippi dismissed the firm’s suit without prejudice: the person impersonated “is not and has never been Gore’s client.” Walk your own scenario through that wording.
Would the Gore endorsement have paid you?
Did the instruction to move money arrive by email?
Did it pretend to come from an employee, or from a person or company already doing business with you for a fee?
Did your employee act in good faith, and did the money leave your control?
Applies only the definition quoted in the March 31, 2026 opinion, under Mississippi law. Other endorsements word it differently.
Travelers’ specimen ties the same cover to an instruction purportedly from a vendor, client or employee, a client being anyone you serve for a fee. No court I found has tested that against a fake new client. If I signed strangers by email, this is the definition I’d ask to have reworded, in writing.
While intake documents are open, the AI clause in your engagement letter can go into the same revision.
Cyber endorsement or standalone policy: sublimits and defense inside the limit
In 2025, 62.8 percent of US cyber policies in force were endorsements bolted onto another commercial policy, according to the NAIC. Those endorsements brought in 3.9 percent of the premium. The phrase “we have cyber” often describes a small add-on whose limit nobody has looked up.
Cyber endorsements: most of the policies, almost none of the premium
NAIC, 2026 Report on the Cybersecurity Insurance Market, figures 4 and 5, 2025 data.
In a small firm, the kind of team we build for, that endorsement may be the whole cyber program. Check its limit first.
Then check defense costs. Travelers’ application warns the limit may be completely exhausted by amounts paid as defense costs, and its form ends the duty to defend once the limit is gone. That legend gets more of my attention than the premium.
What defense inside the limit leaves for a settlement
Example figures. Retention ignored for simplicity. Some policies pay defense outside the limit; your declarations page says which.
Can a cyber claim be denied for no MFA? What the application promises
The Travelers renewal application, CYB-14202, asks about multifactor authentication three separate times: privileged access, remote access to the network, and remote access to email. It also asks whether critical patches go in within 30 days and whether backups are tested annually. The signer states the answers may be relied upon.
In July 2022 Travelers asked a federal court in Illinois to rescind a cyber policy after a ransomware attack, alleging International Control Services had used MFA only on its firewall. I could confirm the complaint, not the outcome. Nobody should tick these from memory; if an outside IT provider runs your systems, they answer.
Could you answer yes to these and prove it tomorrow?
Questions as asked on Travelers CYB-14202. Tick only what you could show evidence for.
0 of 7 you could prove
Every box you left empty is an answer to fix before you sign, or to answer as no.
Keep the evidence with the signed application, dated.
A checklist for preparing answers, not a statement of what any insurer will accept.
One carrier, At-Bay, also lists unapproved or unlicensed software among things a policy may not cover. A one-page list of approved tools is the cheapest answer to that, and an AI use policy template gets you most of the way.
For a US firm, the state-law clauses in our US template belong in the same document.
Non-affirmative cyber risk: what “silent cyber” means for your other policies
New York’s Department of Financial Services described silent risk in 2021 as paying cyber losses under a policy that never mentions cyber, and told insurers to end it by stating whether each policy covers them. The NAIC’s 2026 report calls it non-affirmative coverage and says gaps persist across property, D&O and general liability.
For a buyer, silence cuts both ways. These two cases show it:
| Case | Policy | What happened | What it tells you |
|---|---|---|---|
| EMOI Services v. Owners Insurance, Ohio Supreme Court, Dec. 27, 2022 | Business owner’s policy with an electronic equipment endorsement | Ransomware damage to software was not covered: no direct physical loss to the media holding it | A property-style policy may not pay for an attack that breaks nothing physical |
| Merck v. ACE American, N.J. Appellate Division, May 1, 2023 | All-risk property policies | A hostile or warlike action exclusion did not bar the NotPetya losses; reported settled in January 2024 before the state supreme court ruled | Old wording can surprise insurers too, which is why it is being rewritten |
I would not budget on a silent policy paying. Insurers are rewriting those clauses, and the same silence is now appearing around AI, which the questionnaire guide covers under silent AI.
War exclusion and state-backed cyber attacks: who decides attribution
The NAIC report expects war exclusions, widely used in cyber policies since 2023, to contain part of the losses from state-backed operations. It also notes the hard part: proving a link between an attack and a government, at machine speed. It mentions demand linked to the Iran-U.S. conflict that began in February 2026.
| Wording | What it excludes | Who decides it was a state |
|---|---|---|
| Travelers CYB-16001 (2019) | War, declared or not, and military action; a computer attack is excluded only if it supports one | Not set out in the form |
| Lloyd’s Market Bulletin Y5381 | Since March 31, 2023, standalone Lloyd’s cyber policies must also exclude state-backed attacks that significantly impair a state | The clause must state how the parties agree on attribution |
| Merck’s property policies | Hostile or warlike action | The court required military action, so the question never arose |
Lloyd’s is not a US regulator, but alien surplus lines carriers wrote 21 percent of US cyber premium in 2025. On a renewal call, the attribution sentence is the one I would ask to hear aloud. Foreign rules reach US buyers elsewhere too, as the EU AI Act’s reach outside Europe shows.
Ransom payments, regulatory fines and card penalties
Treasury’s Office of Foreign Assets Control warned in September 2021 that paying a ransom to a sanctioned party can be a violation on a strict liability basis, and that license requests get a presumption of denial. New York’s regulator recommends against paying. Policies that cover ransom usually require the insurer’s consent first.
| Cost | What the specimen form says | The outside rule |
|---|---|---|
| Ransom | Travelers pays it as a cyber extortion cost, with prior written consent | OFAC sanctions risk; NY DFS recommends against paying |
| Regulatory fines | Excluded from loss, except civil fines and penalties in a regulatory proceeding, where insurable | The FTC lists fines among costs to ask about; whether one is insurable depends on state law |
| Card brand assessments | Travelers carves back Payment Card Contract Penalties, with their own limit inside the privacy limit | P.F. Chang’s was denied coverage for $1.93 million of assessments in 2016 under a contractual liability exclusion |
| Restitution and disgorgement | Excluded | None found |
In P.F. Chang’s, decided in Arizona on May 26, 2016, the costs came through the restaurant’s merchant agreement. If I took cards, I’d ask for a payment card carve-back by name. Regulatory exposure starts with whether an incident is a breach at all, which our guide on pasting client data into AI works through.
Surplus lines cyber insurance: no guaranty fund if the insurer fails
US cyber premium in 2025 split 51 percent domestic surplus lines, 28 percent domestic admitted and 21 percent alien surplus lines, per the NAIC. The NAIC states that state guaranty funds, which pay claims when an admitted insurer becomes insolvent, are not available in the surplus lines market.
Where US cyber premium was written in 2025
NAIC, 2026 Report on the Cybersecurity Insurance Market, figure 7. Guaranty fund rule: NAIC, Surplus Lines.
The NAIC says surplus lines insolvencies are historically rare, and that market writes risks admitted carriers will not. Rare still isn’t never: I want to know which kind of carrier I hold before an incident. Agents who handle those claim files have their own exposure, which is why we wrote a page for insurance agencies.
Does cyber insurance cover AI incidents? Two policies moving apart
Coalition’s April 2025 policy form added cover for AI-related security events, including deepfake-enabled fraud. Meanwhile Verisk’s ISO introduced optional generative AI exclusions for general liability from January 2026: CG 40 47 removes Coverages A and B, CG 40 48 only Coverage B. Fenwick still calls cyber the steadiest place for AI risk.
Before answering an insurer’s question about AI tools, know what each vendor does with what your staff type. My note to the broker would name each tool, because “AI” is not one product.
| Tool or route | What the vendor says about your data |
|---|---|
| ChatGPT | who at OpenAI and in your workspace can read chats |
| Gemini | Google’s own warning about human reviewers |
| Claude | Anthropic’s training default and its exceptions |
| Copilot | four Microsoft products with four answers |
| Your own apps on a model API | how data leaves through code; Nonimo Gateway redacts it on your servers |
The NAIC’s AI model bulletin, adopted in December 2023, governs how insurers use AI, not whether your use is covered. That answer lives in your endorsement schedule, and the full list of our AI tool guides is there for the follow-up questions.
![A $158,425 wire turned into [AMOUNT_1], the loss a law firm's social engineering claim failed to recover in Gore v. Spinnaker](/guias/en-us/cyber-insurance-social-engineering-wire-tag.webp "The wire in Gore v. Spinnaker")
Questions to send your broker before you renew
The NAIC records an eighth straight quarter of US cyber rate cuts through early 2026, while the loss ratio rose to 53. A cheaper renewal tells you nothing about the wording. Send the broker these eight, and tick each one when the answer comes back in writing with a form number.
Eight questions for the broker
Tick a question only once you hold a written answer that names the form or endorsement.
0 of 8 answered in writing
Chase the unticked ones before the binder is signed.
File the broker's replies next to the declarations page.
Each question comes from a section of this guide. A broker's email is not the policy: check it against the forms it names.
None of this tells you which policy to buy. I build privacy software in Spain, Nonimo for US teams, not insurance; a producer licensed by your state answers for your business. Treat this page as a reading guide to policy wording, not insurance or legal advice.
A breach notice or a claim file is full of client names. If your team drafts either with an AI chat, Nonimo redacts names and IDs on the computer first, so less personal data reaches the vendor.
Your text doesn’t reach Nonimo either; our security page has the detail.
Sources
- NAIC, 2026 Report on the Cybersecurity Insurance Market. 2025 premium, policy and claims data, endorsement share, market split, loss ratio, war and silent cyber sections.
- NAIC, Surplus Lines. No guaranty fund in the surplus lines market; low historical insolvency rate.
- Travelers, CyberRisk coverage form CYB-16001 Ed. 01-19. Exclusions, definition of Loss, Social Engineering Fraud, Client, ransom and limits.
- Travelers, CyberRisk Renewal Application CYB-14202 Ed. 01-19. The MFA, patching, backup questions and defense-within-limits legend.
- Hiscox, Cyber Coverage Part CYBCL-CYB P0001A CW (10/19), specimen. Exclusions and cyber crime coverage terms.
- U.S. District Court, N.D. Mississippi, Gore, Kilpatrick & Dambrino v. Spinnaker Insurance, opinion of March 31, 2026. The facts and the social engineering holding.
- U.S. District Court, D. Arizona, P.F. Chang’s China Bistro v. Federal Insurance, May 26, 2016. Card assessments and the contractual liability exclusion.
- New York DFS, Insurance Circular Letter No. 2 (2021), Cyber Insurance Risk Framework. Silent risk and the recommendation against ransom payments.
- U.S. Treasury, OFAC, Updated Advisory on Sanctions Risks for Facilitating Ransomware Payments, September 21, 2021. Strict liability and the presumption of denial.
- Federal Trade Commission, Cyber Insurance. Fees, fines and penalties among the first-party costs to ask about.
- Lloyd’s, Market Bulletin Y5381, State backed cyber-attack exclusions. Requirements from March 31, 2023.
- Simpson Thacher, summary of EMOI Services v. Owners Insurance and of Merck v. ACE American. Holdings and policy wording.
- Insurance Journal, Travelers Wants Out of Contract With Insured That Allegedly Misrepresented MFA Use, July 12, 2022, and Merck Settles Coverage Dispute With Insurers, January 5, 2024. The ICS complaint and the Merck settlement.
- Coalition, Active Cyber Policy announcement, April 9, 2025, and 2026 Cyber Claims Report. Base form changes, AI cover, 58 percent and $112,000.
- Chubb, Cyber insurance products. Crime coverages by endorsement.
- Independent Insurance Agents & Brokers of America, Verisk generative AI exclusions. CG 40 47, CG 40 48 and CG 35 08, optional, January 2026.
- Fenwick, The End of Silent AI?, June 15, 2026. Cyber as the steadiest cover for AI risk.
- At-Bay, What is not covered by cyber insurance?. Unapproved or unlicensed software.
Checked on November 7, 2026.
Common questions
What does cyber insurance not cover?
Usually war, your own fraud, bodily injury, physical damage, utility outages and problems known beforehand. Money wired to a fraudster often falls outside the base policy or under a small social engineering sublimit.
Does cyber insurance cover ransomware?
A dedicated cyber policy often does; a property-style policy may not. Travelers' specimen pays ransom as an extortion cost with prior written consent. Ohio's top court found a business owner's policy missed ransomware damage.
What is a war exclusion in a cyber policy?
It removes losses caused by war or military action, declared or not. Since March 2023, Lloyd's standalone cyber policies must also exclude state-backed attacks that significantly impair a state, and say how attribution is agreed.
Can cyber insurance deny a claim over missing MFA?
It can try, if your application claimed MFA you didn't have. In 2022 Travelers asked an Illinois federal court to rescind a policy after ransomware, alleging the insured used MFA only on its firewall.
Does cyber insurance actually pay out?
Yes, when the loss fits a coverage clause and exceeds the retention. A court dismissed one law firm's $158,425 wire claim because its endorsement covered impostors of existing clients, not a fraudster posing as a new one.
Is non-affirmative cyber risk the same as silent cyber?
Yes, they are two names for one gap. Both describe cyber losses that a property, liability or other non-cyber policy neither clearly covers nor clearly excludes.