Cyber insurance renewal in 2026: what changed on carrier applications

SecValley is not an insurance broker, an agent, a carrier, or a law firm, and holds no producer licence. This page describes publicly reported market conditions and how application questions are commonly framed; it is not insurance advice, not legal advice, and not a coverage determination. Your own quote, wording, and renewal outcome depend on your carrier, your broker, and your loss history.

What the market actually did

Marsh's Global Insurance Market Index for the second quarter of 2026 put global commercial insurance rates down 6 percent, with cyber down 4 percent after a 5 percent fall in the first quarter. That is twelve consecutive quarters of cyber declines. The fall was uneven by region: down 14 percent across India, the Middle East and Africa, down 10 percent in Latin America and the Caribbean, down 6 percent in the Pacific, and down 2 percent in the US. Marsh attributed the decline to stable capacity and continuing high levels of insurer competition.

The US being the shallowest decline is the number that matters most to a US buyer. Competition is real, and it is spent on price, limit, and occasionally on a retention. It is rarely spent on the control questions, because those are what a carrier relies on to price the risk at all.

Why the questions moved: what claims did

Coalition's 2026 Cyber Claims Report, covering full-year 2025, is the clearest public read on where losses landed. Initial ransom demands rose 47 percent year over year, while a record 86 percent of affected businesses refused to pay. Business email compromise and funds transfer fraud together accounted for 58 percent of the incidents reported, and 52 percent of funds transfer fraud claims began as business email compromise. Ransomware remained the most expensive claim type, with an average loss of 269,000 dollars, and the average business email compromise loss rose 28 percent to 27,000 dollars.

Read as an underwriting signal, that is a short list. Losses arrive through email and through the ability to restore, and the refusal rate rising to 86 percent is itself evidence that backups and rehearsed response now decide the size of a ransomware loss. Every change described below is downstream of those two facts.

The identity question got sharper

Cisco Talos found authentication abuse in 65 percent of the incidents its response team handled in the second quarter of 2026, up from 35 percent the quarter before, with attackers bypassing or defeating multi-factor authentication through adversary-in-the-middle proxies, stolen session tokens, MFA fatigue, and devices they enrolled themselves.

That is why a bare yes has lost most of its underwriting value: nearly every applicant now says yes. The distinctions that carry information are the ones a 2026 form reaches for.

• Which method is enforced, and whether any part of the estate still accepts a code by SMS or a phone call.
• Which account populations it covers: all mailboxes, remote access, privileged and administrative accounts, service accounts, and external collaborators.
• What the standing exceptions are, counted rather than described. One permanently excluded administrator is the answer, not a footnote to it.
• Whether legacy authentication protocols that bypass the policy are switched off.
• Whether a stolen session token is constrained by anything, such as device compliance or sign-in risk.

Answer these as ratios and method names. "Enforced on 214 of 214 mailboxes, phishing-resistant for the 6 global administrators, legacy authentication blocked tenant-wide, 2 documented exceptions with a remediation date" is an answer a carrier can underwrite and a claims adjuster cannot later reread as an overstatement.

The backup question is now a restore-test question

Immutability has been asked for several years. What tightened in 2026 is the follow-up: the date of the last successful restore test, what was restored, and how long it took. With 86 percent of businesses refusing to pay a ransom, a carrier's exposure on a ransomware claim is essentially the cost and duration of the recovery, so the restore test is the control that prices the claim. A copy that exists but has never been restored from is not a recovery plan, and a 2026 form increasingly says so.

Two further sub-questions travel with it: whether the backup system's own administrative access is separate from the production directory, and whether the retention period outlasts the dwell time of an intrusion. Both are asked because both were how recoveries failed.

Endpoints: a percentage, not a product name

Naming an endpoint detection product answers almost nothing, and applications have moved to the number: what percentage of endpoints have the agent deployed, whether servers are included, and whether alerts are monitored by someone outside business hours. Coverage percentage is readable from the environment; monitoring is attested. Keep the two apart when answering, because they have different evidence.

Funds transfer: the question follows the loss

With business email compromise and funds transfer fraud at 58 percent of incidents, the payment-process questions are no longer a formality at the end of the form. Expect out-of-band verification of any change to payee bank details, a named threshold above which a second approver is required, and whether the verification uses a number already on file rather than one supplied in the request. The sublimits attached to these covers are small enough that carriers contest them, which makes an accurate answer here cheap insurance against a fight later.

The new block: generative AI

The visible change of 2026 sits in general liability rather than cyber. Verisk's ISO filed three generative AI exclusion endorsements for commercial general liability, CG 40 47, CG 40 48 and CG 35 08, effective January 2026. The broad form removes bodily injury, property damage, and personal and advertising injury arising out of generative artificial intelligence; the narrow variant removes personal and advertising injury alone.

Cyber and technology errors and omissions forms are moving more quietly, through appetite questions and endorsement wording rather than a single filed exclusion. What that means practically at renewal is that an inventory of the AI tools in use, and of what data reaches them, is now part of a submission rather than an afterthought. Whether a given policy would exclude an AI-related loss is a wording question for coverage counsel, not a question this page can answer.

Which 2026 answers you can verify, and which stay attested

The practical division at renewal is not between hard and easy questions. It is between answers that can be read from a live environment on a dated basis and answers that rest on a person's statement. Both are acceptable. They just need to be labelled differently in your own preparation, because an attested answer is the one that needs a named owner and a date.

What to do before the next renewal

Nothing in the list above requires a project. It requires the answers to be read rather than remembered.

• Retrieve last year's completed application and supplemental. It is the only document that tells you what you already represented.
• Re-read each prior answer against the live environment, and write the number rather than the yes. Where the reading disagrees with last year's answer, that is the finding.
• Give every attested answer a named owner and a date. An attested answer without one is the answer that fails later.
• Where a gap is real, let the broker carry it into the negotiation with a remediation date attached. An accurate no with a plan prices better than a yes that a forensic report contradicts.

The renewal readiness checklist lays that out as a 90-day sequence. The question library works through the individual questions, what the carrier is testing with each, and where the evidence lives.

Frequently asked

Did cyber insurance get cheaper in 2026?

On average yes. Marsh recorded cyber rates down 4 percent globally in the second quarter of 2026, the twelfth consecutive quarter of declines, with the US down 2 percent and the widest fall, 14 percent, across India, the Middle East and Africa. Overall commercial rates fell 6 percent in the same quarter. An average is not a quote: a risk with a weak control answer or a recent loss can still renew flat or up.

If rates are falling, why is the application harder?

Because price and underwriting moved in different directions. Capacity and competition pushed rates down, while claims kept concentrating on a small number of controls. Carriers therefore compete on price and limit rather than on the control questions, and the questions themselves became narrower and more evidential: a method rather than a yes, a percentage rather than a product name, a date rather than a policy document.

Does answering yes to multi-factor authentication still satisfy the MFA question?

Less often than it did. Cisco Talos found authentication abuse in 65 percent of the incidents it responded to in the second quarter of 2026, up from 35 percent the quarter before, with attackers defeating MFA through adversary-in-the-middle proxies, stolen session tokens, push fatigue and self-enrolled devices. A yes covering 90 percent of accounts with SMS codes describes a different risk than a phishing-resistant method with no standing exclusions, and the 2026 form is written to tell them apart.

Is there a new artificial intelligence question on insurance applications in 2026?

There is new AI wording in the market. Verisk's ISO filed three generative AI exclusion endorsements for commercial general liability, CG 40 47, CG 40 48 and CG 35 08, effective January 2026. Those are general liability forms rather than cyber forms, but they are the clearest indication of direction, and submissions now more often ask which AI tools are in use and what data reaches them. Whether a particular policy excludes an AI-related loss is a wording question for coverage counsel.

Which 2026 answers can be verified from the environment?

Most of the identity and endpoint block: MFA method and coverage per account, legacy authentication, standing exceptions, privileged account inventory, EDR deployment percentage, mailbox auditing, email authentication records, and cloud backup immutability and retention. Restore-test dates, out-of-hours monitoring, payment callback procedures and prior incident knowledge remain attested, which is why each needs a named owner and a date.

Should a softer market change how a renewal is prepared?

Only in what it is worth asking for. The preparation is the same, and it is worth more in a soft market, not less: a submission whose control answers are evidenced is what lets a broker ask for the limit increase or the retention reduction the market is currently willing to give.

Sources

Every market and claims figure on this page is drawn from the publications below. Links open on the publisher's own site.

• Global commercial insurance rates fall 6 percent in the second quarter of 2026, Global Insurance Market Index Q2 2026 Marsh
• Abundant capacity and competition drive 6 percent global commercial insurance rate decline in Q2 2026 Reinsurance News
• 2026 Cyber Claims Report: initial ransom demands surged 47 percent but most businesses refused to pay (5 March 2026) Coalition
• IR Trends Q2 2026: phishing and weaponized remote management tools drive attack chains Cisco Talos
• Verisk to roll out new general liability exclusions for generative AI exposures Independent Insurance Agents & Brokers of America, Virtual University
• ISO introduces generative AI exclusion in commercial general liability policies Gallagher

Sources are cited only for the facts attributed to them. The publishers listed are unaffiliated with Insurance Posture and SecValley, have not reviewed or endorsed this page, and their inclusion implies no relationship. Rate movements are published averages across a broker's portfolio, not quotes, and individual renewals vary widely. Statements about how application questions are commonly framed are editorial assessments and are labelled as such above.

The questions behind this page

Related reading

Back to resources