Cyber Insurance and AI: The Questions Appearing on Applications

AI questions are appearing on a growing number of cyber insurance applications and AI supplements. Practice varies by carrier and by line, and there is no single standard question set. Below is a guide to the questions that are documented as appearing, what each one is really asking for, and how to prepare an answer you can substantiate later.

A note on sourcing before you read further. Carriers rarely publish their application wording, so most of the questions below are representative: they reflect themes documented in trade press and in carrier and broker commentary, rather than text lifted from a named carrier’s form. Where a question is quoted from a real document or attributed to a named individual, this article says so. The Cyber Insurance & AI Tracker is the maintained record with per-item sourcing, and it is the better reference if you need to cite something.

Answer against the application actually in front of you, not against this list.

Why Underwriters Care About AI

The underwriter cares about loss probability, the likelihood that an incident triggers a claim. AI use affects loss probability in several ways:

1. AI tools are a new data exfiltration vector

Sensitive data leaving the organization through AI tool pastes does not show up in traditional DLP, EDR, or CASB monitoring. From a loss probability standpoint, AI tool use is a meaningful exfiltration channel that most organizations have not yet covered.

2. AI-assisted phishing is rising

Generative AI lowers the cost and improves the quality of social engineering attacks. Underwriters increasingly factor AI-assisted attack likelihood into their pricing models, and they want to see that the insured has corresponding defenses.

3. AI use creates new compliance exposure

For underwriters offering regulatory defense coverage, AI use creates new compliance exposure. Regulatory inquiries about AI tool use are increasing across SOC 2 auditors, HHS, EU data protection authorities, and state attorneys general.

4. AI-generated incidents are emerging

A small but growing category of incidents directly involves AI tools, credential leakage to AI vendors, intellectual property exposure, compliance violations from AI tool processing. Underwriters track these incidents and price accordingly.

The Questions Underwriters Are Asking in 2026

Questions of this shape have been reported on 2025 and 2026 applications and AI supplements. Treat the wording as representative rather than verbatim unless noted:

Question 1: “Does your organization have a written AI acceptable use policy?”

What underwriters want to hear: Yes, with a copy available on request. Date of last review. Scope (employees, contractors, devices).

The minimal acceptable answer: A written policy that addresses AI tools specifically, with employee acknowledgement. See our AI acceptable use policy guide for the structure.

Question 2: “What technical controls prevent unauthorized AI tool use?”

What underwriters want to hear: A technical platform that detects AI tool usage and blocks sensitive data submissions. Coverage that includes personal accounts (not network-only). Evidence available on request.

The minimal acceptable answer: An AI governance platform deployed across managed endpoints with content classifiers enabled. Without a technical control, the answer is uncomfortable.

Question 3: “Do you produce audit logs of AI tool activity?”

What underwriters want to hear: Yes, per-event logs with user, timestamp, tool, content classification, and outcome. Retention of at least 90 days. Exportable on request.

The minimal acceptable answer: Audit logs from your AI governance platform that map to your compliance program. Without audit logs, the underwriter assumes the worst.

Question 4: “Are AI vendors in your vendor risk management program?”

What underwriters want to hear: Yes, OpenAI, Anthropic, Google, Microsoft, and any other AI vendors in formal use are in the vendor inventory with completed risk assessments and current agreements.

The minimal acceptable answer: Documented vendor inventory entries with DPAs or equivalent agreements for each approved AI tool.

Question 5: “How do you train employees on AI policy?”

What underwriters want to hear: AI-specific training delivered to all employees, with acknowledgement records retained.

The minimal acceptable answer: AI policy module within your annual security awareness training, with acknowledgement records.

What underwriters want to hear: Yes, your incident response process addresses AI-specific events including data leakage to AI tools.

The minimal acceptable answer: Your existing IR plan updated to include AI tool events as a category, with named owners and escalation paths.

Question 7 (emerging in 2026): “Have you completed a DPIA or AI risk assessment?”

What underwriters want to hear: Yes, formal AI risk assessment (under GDPR DPIA, NIST AI RMF, or equivalent) completed within the past 12 months.

The minimal acceptable answer: Documented AI risk assessment, even if not in a formal framework template.

What Your Answers Are Actually Worth

We do not have underwriting data, and we are not going to pretend otherwise. What a given set of answers does to a specific quote depends on the carrier, the line, the industry, the rest of the security posture, and the jurisdiction. Anyone offering you a premium-impact table for AI answers is guessing.

What is worth saying is narrower and better supported:

An answer you cannot substantiate is worse than a “no.” Applications are warranties. If an answer cannot be evidenced later, it becomes a coverage argument at exactly the wrong moment. In Ace American Insurance Co. v. Congruity 360 and Trustwave (D.N.J., filed September 15, 2025), a Chubb subsidiary sued an IT provider and an MSSP directly to recover a $500,000 ransomware payout, alleging the controls the policy assumed were not actually enforced. That case is recent and unresolved, but it is the clearest current illustration of where unsubstantiated control answers lead.
“We prohibit it” and “we prevent it” are different answers to the same question, and the difference is technically meaningful. See technical controls on AI tool access.
Some carriers do not ask about AI at all yet. Absence of AI questions on your application is not evidence that AI is irrelevant to your coverage; it may simply be that this carrier has not updated its form.

Two of these answers are easier to defend with evidence than with a policy document. Underwriters increasingly want to know that unapproved tools are actually prevented rather than merely prohibited, which is a question about technical controls on AI tool access, and they want loss history context, which is where the documented shadow AI incident record is useful in a renewal conversation.

The Renewal Preparation Playbook

If your cyber renewal is in the next six months, the priority order:

90 Days Before Renewal

Publish or update the AI acceptable use policy. Use our free template if you need a starting point. The policy is the cheapest single answer.
Add AI vendors to your formal vendor inventory. Document DPAs.

60 Days Before Renewal

Deploy an AI governance platform across managed endpoints. Start in monitor-only mode.
Roll out AI training to employees. Collect acknowledgements.

30 Days Before Renewal

Promote the AI governance platform to blocking on highest-severity classifiers (credentials, PHI for healthcare, EU personal data for GDPR-covered organizations).
Confirm audit logs are producing records and retention is configured.
Update your incident response process to address AI-related events.

Renewal Submission

Answer accurately, not optimistically. The application is a warranty. A “yes” you cannot evidence is a liability, not an advantage.
Have evidence available on request, a sample audit log, the policy document, the vendor inventory entry.
Describe what the control actually does, including its scope and its gaps. “Monitored on managed Windows endpoints via browser extension and endpoint agent; unmanaged and personal devices are out of scope” is a better answer than “yes,” because it is one you can defend.

Most organizations can make meaningful progress on this sequence in 90 days. What that is worth at renewal is between you, your broker, and your carrier.

How AI Governance Affects Specific Coverage Lines

The pattern varies by coverage:

Cyber liability (data breach coverage)

AI tools are a plausible data-egress path, so controls over them are relevant to the same risk this coverage responds to. Whether a given carrier prices that relevance, and how, is a question for your broker.

Regulatory defense

Where regulatory defense is covered, the practical value of AI records is in the defense posture: being able to show what was configured, when, and what was detected. That is an evidentiary benefit rather than a demonstrated pricing one.

Business interruption

Less directly affected, but AI-related ransomware variants and AI-assisted attacks contribute to business interruption risk. AI governance is mentioned in some BI questionnaires.

AI-assisted phishing is now standard in attacker toolkits. Underwriters increasingly want to see that the insured has corresponding defenses including employee awareness training on AI-assisted threats.

What the Market Commentary Actually Says

Broker and carrier commentary on AI in cyber is worth reading, but it is commentary, not policy language. Two things are documented and worth separating from the forecasting:

Documented: cyber carriers have been adding affirmative AI coverage (Coalition’s Affirmative AI and Deepfake Response endorsements, AXA XL’s genAI endorsement, the Google Cloud Risk Protection Program with Beazley, Chubb and Munich Re), while general liability has moved the other way, with Verisk/ISO issuing exclusion forms CG 40 47, CG 40 48 and CG 35 08 effective January 1, 2026. Those are different lines moving in opposite directions, and conflating them is the most common error in this topic.

Forecast, not fact: brokers frequently draw an analogy between AI governance and the arrival of MFA requirements, first optional, then expected, then required. It is a reasonable read of the direction of travel. Whether AI follows the same path is not yet established, and it should be labelled as a prediction when you repeat it.

Both are tracked with per-item sourcing in the Cyber Insurance & AI Tracker.

Frequently Asked Questions

What does cyber insurance ask about AI in 2026?

The questions above are the themes that recur: written policy, technical controls, audit logs, vendor inventory, employee training, incident response, and AI risk assessment. There is no standard question set. Carriers write their own applications and AI supplements, and some do not ask about AI at all yet.

How much can AI governance affect cyber insurance premiums?

We do not know, and neither does anyone publishing a number without underwriting data behind it. Pricing depends on the carrier, the line, the industry, the overall security posture and the jurisdiction. Ask your broker what it is worth on your specific renewal.

Coverage varies by carrier and policy. The clearest documented movement so far has been in general liability rather than cyber: Verisk/ISO made exclusion forms CG 40 47, CG 40 48 and CG 35 08 available effective January 1, 2026, and carriers choose individually whether to adopt them. Several cyber carriers have moved in the opposite direction and added affirmative AI language. Read your actual policy, and ask your broker to confirm per line and in writing.

What if our renewal is in two weeks and we have no AI controls?

Answer accurately and start the controls work. Some carriers will consider a documented remediation plan. What is not advisable is claiming a control you cannot evidence, because the application is a warranty and the claim is where that gets tested.

Do underwriters care which AI governance platform you use?

We have no evidence that carriers recognise or prefer specific AI governance vendors by name, and you should be sceptical of any vendor telling you otherwise. What matters is whether you can describe what the control does, what it covers, and what it does not, and produce records to support that description.

How long does it take to go from no controls to renewal-ready?

Deploying endpoint tooling across a managed fleet is typically a two to three week project. Policy, training records and an incident-response update are organizational work that runs in parallel. “Renewal-ready” depends on what your carrier actually asks.

What about cyber insurance for MSPs?

MSPs face this from both sides: their own policy, and the answers they help clients give. The subrogation exposure is the part worth attention, since Ace American v. Congruity 360 and Trustwave names an IT provider and an MSSP directly. See our MSP AI governance service guide for the packaging pattern, and the Underwriter Field Guide for what an MSP can and cannot reasonably attest to.

Cyber insurance is one of the more concrete pressures pushing AI governance up the priority list, because a renewal has a date on it. The useful goal is not to maximise the number of “yes” answers. It is to be able to describe your AI controls accurately, including their limits, and to have records that support the description if anyone asks later.

Nothing here is legal, insurance or coverage advice. Coverage varies by carrier, policy and jurisdiction; verify against your actual policy wording and your broker.