Sometimes. Underwriting files show how an underwriter read a risk, priced or declined it, and how the account later performed, which is rare training data for AI underwriting assistants. Mercor lists underwriting files among its target data. The limits are strict: the Gramm-Leach-Bliley Act and every state's version of NAIC Model 672 restrict sharing personal information, health information needs the person's authorization, and in practice personal and health details never go into a license. Commercial lines are the easier start.
- The useful unit is one submission: the risk, the underwriter's reasoning, the decision, and the loss result years later.
- Commercial lines files about businesses are easier to license than personal lines files about individuals.
- GLBA and state rules based on NAIC Model 672 limit disclosure of nonpublic personal information; health information needs authorization.
- This page covers underwriting. Claim files are covered in the insurance claims guide.
Can an insurer, MGA or agency license underwriting files to AI companies?
Sometimes, and mostly for commercial business. An underwriting file records how a professional judged a risk: what was submitted, what they asked for, the price or the reason for declining, and the conditions attached. Linked to how the policy later performed, it becomes a decision with a measured outcome.
Company data programs name it. Mercor's company data page lists "Finance & Insurance Statements, underwriting files, claims" among its target industries (Mercor). Springbolt Group, an advisory firm, counts "Actuarial calculations with underlying assumptions" among specialist documents it says labs pay a premium for (Springbolt Group).
How underwriting files differ from claims data
Claims data records what happened after a loss: notice, investigation, reserves and payment. It is covered in selling insurance claims data to AI companies. Underwriting files record the decision before the policy exists, and their outcome is measured later through premiums and losses.
| Underwriting files (this page) | Claims files | |
|---|---|---|
| When | Before and at binding or renewal | After a loss is reported |
| Core decision | Accept, price, set terms or decline | Cover or deny, reserve, settle |
| Outcome that labels it | Loss ratio and claims on the policy in later years | Final payment and closure |
| Who usually holds it | Carriers, MGAs with delegated authority, wholesalers and agencies | Carriers, TPAs, adjusters |
| Main extra issue | Declined applicants never became customers but are still in the file | Injury and health details in the claim |
What an AI buyer needs: the decision, its context and the recorded outcome
A complete underwriting example has all three parts:
- Raises value: underwriter notes in plain language, declinations with reasons, and loss data you can join to the original file.
- Raises value: several years and several lines of business, with consistent class codes.
- Lowers value: rating outputs with no notes, bound policies with no link to later losses, or files with third-party reports you cannot share.
| Decision | Context the buyer needs | Recorded outcome |
|---|---|---|
| Quote or decline a new submission | Application, loss runs, inspection report, appetite guide, underwriter notes | Bound or not; if bound, losses over the following years |
| Set price and deductible | Rating worksheet, credits and debits applied and why | Premium earned against losses incurred |
| Add an exclusion or condition | The risk feature that prompted it | Whether a claim later tested that wording |
| Renew, non-renew or re-price | Prior losses, changes in the risk, market notes | Retention and later loss experience |
| Refer above authority | Reason for referral, reviewer's comments | Final decision and result |
What AI companies pay for insurance records
No buyer publishes a price for underwriting files alone. The published ranges cover a company's whole data license:
| Program | Published range or rule | Eligibility stated |
|---|---|---|
| micro1 Company Data Partnerships | Payments "typically ranging from $100K to over $3M", paid within 30 days of acceptance (micro1) | Established companies with 30+ full-time employees and substantial historical operational data |
| Handshake AI data partnerships | "$100K to $4M+ payouts", described as an indicative estimate, not an offer (Handshake AI) | 20+ full-time employees and 3+ years in operation |
| Mercor data monetization | No range published; compensation "is based on data volume, the tools you connect, and the depth of your data" (Mercor) | Target industries listed on the page |
These are company-wide ranges for all the records a company licenses, not prices for one record type, and no program publishes a price per record for the data on this page. Read them as the size of deals programs say they make, not as a quote.
GLBA, the NAIC model privacy rules and state insurance departments
Insurance privacy is regulated by federal law and enforced by the states. What each rule says:
- GLBA notice. A financial institution "may not, directly or through any affiliate, disclose to a nonaffiliated third party any nonpublic personal information" unless it has given the consumer a privacy notice (15 U.S.C. 6802).
- GLBA opt-out. Before such a disclosure, "the consumer is given the opportunity, before the time that such information is initially disclosed, to direct that such information not be disclosed to such third party" (15 U.S.C. 6802). An AI developer is a nonaffiliated third party.
- State enforcement. For insurers, GLBA privacy rules are enforced "in the case of any person engaged in providing insurance, by the applicable State insurance authority of the State in which the person is domiciled" (15 U.S.C. 6805), and states may go further where their protection "is greater than the protection provided under this subchapter" (15 U.S.C. 6807).
- NAIC models. The NAIC lists "The Privacy of Consumer Financial and Health Information Regulation (#672)" and "The NAIC Insurance Information and Privacy Protection Model Act (#670)" among its privacy models, and states: "Every state has adopted #672 to comply with Gramm-Leach-Bliley Act requirements." It also notes a working group "is currently drafting amendments to modernize Model #672." (NAIC)
- A state example. Washington's insurance rule, "Chapter 284-04 WAC PRIVACY OF CONSUMER FINANCIAL AND HEALTH INFORMATION", says a licensee "may not, directly or through any affiliate, disclose any nonpublic personal financial information about a consumer to a nonaffiliated third party unless" notice and opt-out conditions are met. Its definition of personally identifiable financial information covers information "A consumer provides to a licensee to obtain an insurance product or service from the licensee", which is what an application is (WAC 284-04).
- Health information. The same Washington rule says: "A licensee shall not disclose nonpublic personal health information about a consumer or customer unless an authorization is obtained from the consumer or customer whose nonpublic personal health information is sought to be disclosed." (WAC 284-04) Life, disability, health and workers' compensation files are full of it.
What this means in practice: personal and health information never goes into a license. Names, addresses, dates of birth, policy numbers, driving and credit details, and any medical information are removed before a file is considered, and personal lines files are much harder to use at all. Your state may add its own rules, and your domiciliary insurance department enforces them.
This page states what each rule says, with its source. It is not legal advice: your counsel decides how the rules apply to your company and your records.
Other rights checks in an underwriting file
Beyond privacy law, an underwriting file holds material that belongs to others:
- Third-party reports. Credit-based scores, motor vehicle records, loss history databases, inspection reports and property data are bought from vendors under licenses. Check whether yours allow any redistribution; if not, leave the reports out and keep only your underwriter's summary.
- Delegated authority. An MGA or program administrator writing on a carrier's paper should check its agency agreement: the carrier may own or control the underwriting files.
- Agencies and brokers. Submission documents belong to the insured and may be confidential; a retail agency's notes are usually its own, but its carrier and client agreements decide.
- Rating plans. Internal pricing models and appetite guides may be trade secrets you choose to keep out.
How to prepare underwriting files for licensing
Commercial lines files with business insureds are the realistic starting point.
- List lines of business, years, systems (policy admin, underwriting workbench, document store) and whether losses can be joined to policies.
- Separate commercial lines from personal lines; set aside anything with health information.
- Remove individual names, contact details, dates of birth, policy and claim numbers, and third-party reports.
- Keep the reasoning: underwriter notes, referral comments, declination reasons and conditions.
- Join each file to its later loss history by a code, not by a policy number.
- See the de-identification guide and GDPR and CCPA in data licensing for method.
Next step
Start with one commercial line where your underwriters write good notes and losses are linked to policies. Claims teams should also read the insurance claims data guide.
Does your company hold records like these? See what your data is worth.
Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.
Common questions
Can an insurance company sell underwriting data to AI companies?
Sometimes, as a de-identified license, and mostly for commercial lines. Mercor lists underwriting files among its target data. GLBA and state rules limit disclosure of nonpublic personal information, and health information needs authorization, so personal and health details are removed.
What is in an underwriting file that AI developers value?
The submission, the underwriter's notes and questions, the price or the reason for declining, any conditions, and the policy's loss results in later years. The reasoning plus the later outcome is the scarce part.
Does GLBA apply to sharing underwriting files?
It can. GLBA bars a financial institution, including an insurer, from disclosing nonpublic personal information to a nonaffiliated third party without notice and an opt-out. State insurance authorities enforce it for insurers, and every state has adopted NAIC Model 672.
Can health information from underwriting be licensed?
No, not in practice. State rules based on NAIC Model 672, such as Washington's WAC 284-04, bar disclosing nonpublic personal health information without the person's authorization. Leave it out.
How are underwriting files different from claims files?
Underwriting files record the decision before a policy is written, and their outcome is the loss history that follows. Claims files record what happened after a loss. Origin covers claims data in a separate guide.
Sources and further reading
- Mercor: data monetization for companies (target industries and compensation)
- micro1: Company Data Partnerships (payment ranges and eligibility)
- Handshake AI: get paid for your company data
- Springbolt Group: What AI labs actually pay for (October 2026)
- 15 U.S.C. 6802: GLBA limits on disclosure of nonpublic personal information
- 15 U.S.C. 6805: GLBA enforcement, including state insurance authorities
- 15 U.S.C. 6807: relation to state laws giving greater protection
- NAIC: Data privacy and insurance (model laws 670 and 672)
- Washington WAC 284-04: privacy of consumer financial and health information