How to Build an AI Tool Inventory for Cyber Insurance

An AI tool inventory for cyber insurance renewal is a maintained record of every AI service your staff use, sanctioned or not, with an owner, a business purpose, the data each tool receives, the account tier and the vendor's retention and training terms, plus evidence behind every entry. Some insurers and brokers now ask about AI use at renewal, but the questions vary by carrier, and no inventory guarantees coverage. It lets you answer what the form actually asks, accurately.

This guide expands step 1 of our AI questions in cyber insurance renewals checklist. The companion guides cover the AI acceptable-use policy and AI data flows and vendor controls.

Why AI tool questions show up at renewal

The evidence comes from advisers and brokers, not from a survey of carriers. Northstar IT Advisory, a UK consultancy, wrote in October 2026 that the most noticeable addition in many 2026 questionnaires is a section on AI, where underwriters want to know what AI tools the business uses, what data is sent to third-party AI providers, and what controls are in place. ACA Group, writing for US investment managers in July 2026, lists AI tool inventories among the items cyber insurers are increasingly evaluating. Insurance Business reported in August 2026 that there is no standard AI question yet, and that brokers are often expected to volunteer AI use they judge to be material.

Read those as market observations. Your carrier may ask one question, ten, or none. The inventory is useful either way, because it is the record you draw on for whatever the application asks and for any disclosure your broker advises.

Northstar also describes the usual starting position plainly: for many small firms, the honest answer to "what AI tools are in use?" is "we're not entirely sure." Staff use chat assistants through company accounts and personal ones, often without anyone deciding what data is acceptable to put into them. Closing that gap is the job of the inventory.

What should an AI tool inventory include?

Each entry should name the tool, its owner, why the business uses it, which data categories go into it, which account tier or contract applies, and what the vendor's retention and training settings are. Each of those fields should point to evidence rather than memory.

A workable column set:

• Tool and vendor: product name, vendor, and whether it is a standalone service or an AI feature inside a platform you already license.
• Status: sanctioned, under review, or prohibited.
• Business owner: a named person accountable for the tool, not a department.
• Technical owner: whoever administers the tenant, licences and settings.
• Business purpose: one sentence, for example "drafting first replies to customer support tickets".
• Users: the number of users or the groups with access.
• Data categories: what staff are allowed to enter and what they actually enter, using your existing data classification (customer personal data, financial records, health or other regulated data, source code, credentials, legal material).
• Account tier: a company tenant under a business agreement, or individual consumer accounts.
• Retention and training: what the vendor keeps, for how long, and whether your inputs can be used to train its models under your contract and current settings.
• Actions: whether the tool only generates text or can also act, such as writing to production data, approving workflows or sending messages.
• Evidence: links to the contract, the admin settings export, the risk assessment and the approval record.
• Last reviewed: date and reviewer.

NIST's AI Risk Management Framework playbook describes the same discipline for larger programs under GOVERN 1.6: policies that define who maintains the inventory, which systems are included and which attributes are recorded, with a stated preference for inventorying all systems, because partial inventories may not provide the value of a full one.

How do you find unsanctioned AI tools?

Start with the systems that already record what staff connect to and pay for: identity provider app grants, browser extension reports, expense data, and DNS or proxy logs. Then ask teams directly, because logs miss personal devices and personal accounts.

Practical discovery sources:

• SSO and OAuth consent grants in your identity provider, which show third-party apps that staff authorised against company accounts.
• Browser extension reports from your endpoint management tool. Many AI writing and meeting tools ship as extensions.
• Expense and corporate card data for AI subscriptions bought by individuals.
• DNS, proxy or secure web gateway logs filtered for AI service domains.
• Admin consoles of SaaS platforms you already pay for, where AI features may have been switched on by a tenant admin or by a vendor update.
• A short questionnaire to each team lead asking what they use and for what.

Keep rows for tools you decide to block, with their status and the date of the decision. An entry such as "prohibited, found in use by one team, blocked at the web proxy" is better evidence of control than a list that only shows approved tools.

Recording account tier, retention and training settings

The same product can carry different terms depending on the account. A consumer sign-up and an enterprise tenant from one vendor may differ in retention, training use, admin controls and contractual commitments. Record the tier each group actually uses, and keep a dated copy of the relevant settings and contract clause. Vendor terms and defaults change, so date every capture and recheck it at each review instead of relying on what a marketing page said last year.

If you route usage through an LLM gateway or a data loss prevention control, record that in the inventory and keep a policy export and a sample of its logs. Evidence that a control ran is stronger than a statement that one exists.

Evidence to keep for each inventory entry

• The approval record, or the decision to prohibit.
• The risk assessment done before rollout. ACA lists documented pre-deployment risk assessments among the evidence it expects carriers to ask about.
• The contract or terms excerpt covering retention, training use and data location.
• An admin settings export or dated screenshot.
• The user or group list as of the review date.
• For tools that can act on production data, any testing results and a short description of where a human approves the output.

How often should you update the AI tool inventory?

Update an entry whenever a tool is approved, retired or moved to a different account tier, and run a full review before each renewal. A scheduled discovery pass between renewals stops the list drifting out of date.

Northstar's renewal advice applies directly: put last year's form beside this year's, test each answer against reality including the exceptions, and gather evidence before you answer. In its words, "Yes, except for..." needs to be written down, not glossed over. The inventory is where those exceptions live.

What the inventory does not do

An inventory is evidence, not an outcome. It does not guarantee cover, a renewal, a premium or a claim payment, and it does not tell you what your policy wording says about AI. Use it to answer the questions the application actually asks, accurately, and ask your broker how any other AI use should be disclosed.

Disclosure duties differ by jurisdiction. Northstar gives the UK Insurance Act 2015 duty of fair presentation as one example, noting that an answer given in good faith that turns out to be wrong can still affect a claim. Have your broker or counsel review the answers and the policy wording. This article is not legal advice.

If you are building the rest of the evidence pack at the same time, the SMB security pack ten questions cover the baseline controls most applications ask about.

Sources

• Northstar IT Advisory: Your cyber insurance renewal will ask questions your last one did not
• Insurance Business: Your client's staff are pasting data into ChatGPT. Their cyber insurer wants to know about it
• ACA Group: How AI is changing the rules of cyber insurance
• NIST AI RMF Playbook: Govern