meilynx

Reference · Insurance

NYDFS Circular Letter No. 7 on Insurer AI, Explained

Insurance Circular Letter No. 7 (2024): Use of Artificial Intelligence Systems and External Consumer Data and Information Sources in Insurance Underwriting and Pricing

Last reviewed November 15, 2026

At a glance

Issued
11 July 2024, the final form of a proposed circular letter published 17 January 2024
Issued by
New York State Department of Financial Services, Insurance
Addressed to
Insurers authorized in New York, Article 43 corporations, HMOs, licensed fraternal benefit societies and the New York State Insurance Fund
Covers
Underwriting and pricing of insurance policies and annuity contracts
Legal effect
The Department's expectations under the Insurance Law, examined under Insurance Law § 309 or through a special report under § 308
Official text
dfs.ny.gov

Insurance Circular Letter No. 7, issued by the New York State Department of Financial Services on 11 July 2024, sets out what the Department expects of insurers that use artificial intelligence systems (AIS) and external consumer data and information sources (ECDIS) to underwrite and price insurance policies and annuity contracts.DFS Circular Letter 7 · para. 2 It is the final form of a proposed circular letter the Department published on 17 January 2024.DFS Circular Letter 7 · Introduction The Department may audit and examine an insurer's use of AIS and ECDIS in regular or targeted examinations under Insurance Law § 309, or through a request for a special report under § 308.DFS Circular Letter 7 · para. 9

It reaches every insurer authorized in New York, for underwriting and pricing only

The circular is addressed to all insurers authorized to write insurance in New York, Article 43 corporations, health maintenance organizations, licensed fraternal benefit societies and the New York State Insurance Fund. It does not apply to Child Health Plus, the Essential Plan or Medicaid managed care coverage.DFS Circular Letter 7 · para. 2, n. 3

An AIS is any machine-based system that performs functions associated with human intelligence, such as reasoning, learning and self-improvement, and is used in whole or in part to supplement traditional health, life, property or casualty underwriting or pricing, to act as a proxy for it, or to identify lifestyle indicators that feed an underwriting or pricing assessment.DFS Circular Letter 7 · para. 5 ECDIS is data or information used for the same three purposes. It excludes an MIB Group member information exchange, a motor vehicle report, prescription drug data and a criminal history search.DFS Circular Letter 7 · para. 6

The two definitions stand apart. Commenters asked the Department to limit the guidance to AIS that use ECDIS, and the Department kept both definitions so that an AIS is covered whether or not it uses external data.DFS Circular Letter 7 · Introduction

The circular addresses no phase of the insurance product life cycle other than underwriting and pricing.DFS Circular Letter 7 · para. 8 Claims administration and fraud detection, which the NAIC Model Bulletin places inside an insurer's AI program, sit outside it.NAIC Model Bulletin · Section 3, 1.6

External data must relate to risk and must not stand in for a protected class

The starting rule is that an insurer should not use ECDIS or AIS for underwriting or pricing unless it can establish that the data source or model does not use, and is not based in any way on, a class protected under Insurance Law Article 26.DFS Circular Letter 7 · para. 10

Each external data source needs the actuarial support any other rating or underwriting variable needs: generally accepted actuarial standards of practice, actual or reasonably anticipated experience, and an analysis showing a "clear, empirical, statistically significant, rational, and not unfairly discriminatory relationship" between the variable and the insured's risk.DFS Circular Letter 7 · para. 11

The proxy assessment asks a separate question: how far each data source correlates with status in a protected class. The correlation may be measured with data the insurer holds or reasonably inferred with accepted statistical methods, and where one is found, the insurer considers whether a legitimate business necessity requires that source.DFS Circular Letter 7 · para. 12 The Department confirmed that insurers may infer or impute protected class status for this testing and are not expected to collect additional data from or about individuals to do it.DFS Circular Letter 7 · Introduction; para. 18

Testing runs in three steps, and a failed step sends the insurer back to the first

A comprehensive assessment of whether an underwriting or pricing guideline derived from ECDIS or AIS discriminates unfairly or unlawfully includes, at a minimum, three steps taken in order. Step 1 covers every protected class whose membership the insurer can determine from its data or reasonably infer with accepted statistical methods.DFS Circular Letter 7 · para. 15

Testing happens before an AIS goes into production, on a regular cadence after that, and whenever the ECDIS or the AIS changes materially.DFS Circular Letter 7 · para. 17 The insurer documents its testing methods and reasoning in proportion to how it uses ECDIS and AIS and how complex and material they are, and keeps that documentation ready for the Department.DFS Circular Letter 7 · para. 16

Flow

The three-step assessment in paragraph 15

  1. Step 1Disproportionate adverse effect?No prima facie showing: the evaluation can end. A showing: continue to Step 2
  2. Step 2Legitimate, lawful and fair explanation?None: modify the use and restart at Step 1. One exists: continue to Step 3
  3. Step 3Less discriminatory alternative?One exists: modify and restart at Step 1. None: ongoing model risk management, Step 3 repeated at least annually
Steps, order and outcomes from paragraph 15 of Insurance Circular Letter No. 7.

The circular names the statistics, and a qualitative test sits beside them

Insurers are encouraged to use several statistical metrics together on data and model outputs. The circular lists examples, among others:DFS Circular Letter 7 · para. 18

  • Adverse impact ratio: the rates of favorable outcomes for protected classes against control groups.
  • Denials odds ratios: the odds of an adverse decision for protected classes against control groups.
  • Marginal effects: how a marginal change in a predictive variable moves the likelihood of an unfavorable outcome, particularly for members of protected classes.
  • Standardized mean differences: the gap in average outcomes between protected classes and control groups.
  • Z-tests and t-tests: whether those differences are statistically significant.
  • Drivers of disparity: the variables that cause the differences, estimated with methods such as sensitivity analysis, Shapley values or regression coefficients.DFS Circular Letter 7 · para. 18
The statistics do not complete the assessment. The insurer should be able to explain, at all times, how its AIS operates and state a logical relationship between each data source or model variable and the individual's risk.DFS Circular Letter 7 · para. 19

The board owns the outcomes, senior management runs the program, and internal audit tests it

AIS oversight sits inside the corporate governance framework 11 NYCRR § 90.2 already requires.DFS Circular Letter 7 · para. 20 The board, or other governing body, may delegate oversight of ECDIS and AIS to its committees and senior management, with regular reporting that gives it the facts it needs on the material activities and risks.DFS Circular Letter 7 · para. 22 Answering commenters, the Department held the board and senior management responsible for the overall outcomes of ECDIS and AIS use and left the day-to-day implementation with senior management.DFS Circular Letter 7 · Introduction

Senior management sets policies and procedures, assigns competent staff, oversees model risk management, ensures effective challenge and independent risk assessment, reviews internal audit findings and acts promptly on them. It engages a cross-functional committee drawn from legal, compliance, risk management, product development, underwriting, actuarial and data science, as appropriate.DFS Circular Letter 7 · paras. 23, 24 Written policies are reviewed and approved at least annually and cover roles, reporting to senior management and training for relevant staff.DFS Circular Letter 7 · paras. 25 to 28

Internal audit, the function 11 NYCRR § 89.16 requires, is engaged in proportion to the risk. Its review may cover adherence to policy, whether validations happen on time, the accuracy of AIS documentation, the reliability and integrity of ECDIS, potential bias in the data, and whether reporting to the board is enough to judge model risk against appetite.DFS Circular Letter 7 · para. 34

Insurers keep procedures to receive and address consumer complaints about AIS and ECDIS, with the records retained under 11 NYCRR 243.DFS Circular Letter 7 · para. 30 The documentation the Department may ask for includes:DFS Circular Letter 7 · para. 29

  • the process for identifying operational, financial and compliance risks of ECDIS and AIS, and the controls that mitigate them;
  • an up-to-date inventory of every AIS in use, under development, or recently retired;
  • how each AIS operates: its ECDIS and other inputs with their sources, its purpose and products, actual or expected usage, restrictions on use, risks and safeguards;
  • the process for tracking changes, with the rationale for each change and who approved it;
  • the process for monitoring usage and performance, with a list of past exceptions to policy;
  • testing of AIS outputs at least annually, including drift from machine learning or other automated updates;
  • the data lifecycle, from ECDIS acquisition through storage, use and sharing to archival and destruction.

A vendor's system remains the insurer's responsibility

The duty to comply with anti-discrimination law stays with the insurer whether it collects data and underwrites directly or relies on a vendor's ECDIS or AIS as a partial or full substitute. An insurer may not use a vendor to collect or use information it could not collect or use itself, and it may not rely solely on a vendor's claim of non-discrimination or a proprietary vendor process to determine compliance.DFS Circular Letter 7 · para. 14

Oversight runs through written standards for acquiring and relying on vendor ECDIS and AIS, a route for reporting incorrect information to the vendor, and procedures to remove incorrect information from the insurer's own AIS.DFS Circular Letter 7 · para. 36 The Department scaled the diligence to the risk: the insurer performs due diligence and oversight proportionate to the risks of the vendor's ECDIS or AIS, without needing to understand its detailed inner workings, and remains responsible for the outcomes.DFS Circular Letter 7 · Introduction

Where appropriate and available, vendor contracts give the insurer audit rights or audit reports from qualified auditing entities, and require the vendor to cooperate with regulatory inquiries and investigations into the insurer's use of its product.DFS Circular Letter 7 · para. 37 The Department drew that clause from the NAIC Model Bulletin.DFS Circular Letter 7 · Introduction

An adverse decision names its data source, and an ECDIS-based process adds a 15-day notice

Where an insurer uses ECDIS or AIS, it notifies the insured or applicant, or a medical professional they designate. The notice discloses whether the insurer uses AIS in underwriting or pricing and whether it uses data about the person from external vendors, and tells the person they may request the specific data behind the decision, with contact details for that request.DFS Circular Letter 7 · para. 39

For a declination, limitation, rate differential or other adverse underwriting decision, the reasons include all the information the decision was based on and the source of that information. A vendor's proprietary algorithm does not justify a lack of specificity.DFS Circular Letter 7 · paras. 39, 40 Failing to disclose the material elements of an AIS and the external data it relies on may constitute an unfair trade practice under Insurance Law Article 24.DFS Circular Letter 7 · para. 41

Answering requests from life insurers, the circular clarifies the disclosure section of Circular Letter No. 1 (2019), which addressed external data in life insurance underwriting.DFS Circular Letter 7 · para. 42DFS Circular Letter 1 For an underwriting process that uses ECDIS or AIS:

  • Eligibility thresholds for the process, such as age, coverage amount or non-smoker status, are disclosed clearly and prominently in advertisements, marketing materials and application disclosures.DFS Circular Letter 7 · para. 43
  • When the process will not approve an applicant, who can then be covered only through other underwriting, the insurer gives written notice of the reasons within 15 days of that determination, through the channel the applicant chose for communications. The other underwriting process continues meanwhile.DFS Circular Letter 7 · para. 44
  • When specific ECDIS data led to that result, the insurer offers a process to review those data for accuracy, at the time of the notice.DFS Circular Letter 7 · para. 45
  • Failing to give the notice or the review process may be considered an unfair trade practice under Article 24.DFS Circular Letter 7 · paras. 44, 45

New York runs its own rule beside the NAIC bulletin and Colorado

New York has not adopted the NAIC Model Bulletin. The NAIC's implementation map places it with California, Colorado and Texas among the states that govern insurer AI through insurance-specific rules of their own.NAIC map

The three instruments share governance, inventory and vendor oversight. They differ in scope and in what the insurer produces. The NAIC bulletin covers the whole insurance life cycle and is examined through a market conduct request list.NAIC Model Bulletin · Sections 3, 4 Colorado's Regulation 10-1-1 aims its testing at unfair discrimination with respect to race and requires an annual report signed by an officer.3 CCR 702-10 · Sections 5.A, 6 New York's circular stays inside underwriting and pricing and goes furthest on how the testing is done and on what an applicant is told.

PointNY Circular Letter No. 7NAIC Model BulletinColorado Reg 10-1-1
Activities coveredUnderwriting and pricingThe insurance life cycle, from product design and marketing to claims and fraud detectionInsurance practices that use external consumer data, or algorithms and models that use it
WhoInsurers authorized in New York, Article 43 corporations, HMOs, fraternal benefit societies, the State Insurance FundInsurers licensed in an adopting stateInsurers offering individual life, private passenger auto or health benefit plans
Discrimination testingThree steps for every protected class the insurer can determine or infer, with named metricsBias analysis, validation and testing inside the AIS ProgramQuantitative testing for unfair discrimination with respect to race
What the insurer producesDocumentation on the Department's requestThe program and its records in an investigation or market conduct examinationAn annual compliance report signed by an officer, filed in SERFF
Consumer noticeAIS and external data use, the source behind an adverse decision, a 15-day notice from an ECDIS-based processNotice to impacted consumers that AI Systems are in useComplaint and inquiry protocols for consumers acting on an adverse decision
Sources: NYDFS Insurance Circular Letter No. 7 (11 July 2024), the NAIC Model Bulletin (adopted 4 December 2023), and 3 CCR 702-10 as amended effective 15 October 2025. NYDFS Insurance Circular Letter No. 7

Control mapping

What a reviewer expects to be able to see.

ObligationWhat the system must doEvidence a reviewer expects
Actuarial validity and proxy assessment (paras. 11, 12)Show each external data source has a statistically significant, rational relationship to risk, and assess its correlation with protected classesActuarial support per variable, the proxy correlation analysis and any business-necessity rationale
Three-step assessment (paras. 15, 17)Test before production, on a regular cadence and at every material change, and repeat the alternatives search at least annuallyDated results for each step, the alternatives considered and the decision taken
AIS inventory and descriptions (para. 29)Keep an up-to-date inventory of AIS in use, under development and recently retired, with inputs, sources, purpose and restrictionsThe inventory and each system's description, with the date of last update
Change tracking (para. 29)Record each change in ECDIS and AIS use, its rationale and who approved itChange log with rationale and approver
Board and senior management oversight (paras. 21 to 26)Report material AIS activity and risk to the board, and approve policies at least annuallyBoard and committee reporting, and the annual policy approval record
Third-party vendors (paras. 14, 35 to 37)Oversee vendor ECDIS and AIS under written standards, with audit and cooperation terms where appropriate and availableVendor standards, diligence files, contracts, and corrections reported to vendors
Notices (paras. 39, 44, 45)Name the information and its source behind an adverse decision; give the 15-day notice and the accuracy review in an ECDIS or AIS-based processNotice templates, dated notices as issued, review requests and their outcomes

Key dates

Timeline, drawn to scale

From Circular Letter No. 1 to Circular Letter No. 7

In effect as of November 15, 2026Circular Letter No. 7
2020202120222023202418 JAN 2019Circular Letter 114 NOV 2023Colorado 10-1-14 DEC 2023NAIC bulletin17 JAN 2024Proposed11 JUL 2024Circular Letter 7
Dates from Insurance Circular Letter No. 7, Insurance Circular Letter No. 1 (2019), the NAIC Model Bulletin and 3 CCR 702-10. Status as of November 15, 2026.

Sources cited

Common gaps

Where insurers most often fall short of the circular when the Department asks.

  • Testing scoped to external data alone. The Department kept the AIS definition independent of ECDIS, so a model that uses no external data and supplements traditional underwriting is still an AIS. A testing program built around data sources misses it.
  • A vendor's fairness statement in place of testing. Paragraph 14 rules out relying solely on a vendor's claim of non-discrimination or a proprietary process. The insurer needs its own assessment, or the vendor's evidence under a contract that lets the insurer see it.
  • Step 3 run once. Where no less discriminatory alternative exists, the search repeats at least annually. One analysis dated to the model's launch leaves every later year unanswered.
  • Adverse decision letters that give a reason and omit the source. Paragraph 39 asks for the source of the specific information behind an adverse decision. A reason code from a vendor score with no source leaves the applicant nothing to correct.
  • No clock on the ECDIS-based process. The 15-day notice runs from the determination that an applicant cannot be underwritten through the ECDIS or AIS-based process. An organization that tracks only the final underwriting decision has no date to measure it from.
  • Retired systems dropped from the inventory. The inventory covers systems recently retired as well as those in use and under development. A list kept only of live systems cannot answer for a decision a retired model made.

Last reviewed November 15, 2026. This reference summarises publicly available regulatory guidance and is provided for general information. It is not legal advice. Obligations depend on an institution's charter, registration status, size, and activities. Verify against the primary sources cited above and consult counsel before relying on any summary here.

Regulatory updates

When a regulator changes what an AI examination asks for, hear about it first.

Short notes on SR 26-2, NYDFS 500, FINRA, the NAIC bulletin, the EU AI Act, and the employment-AI statutes, plus what we ship. A few emails a month.