When the insured opens a decline letter and sees "based on factors including credit-based insurance score and prior loss history, this decision involved an automated decision system," the insured is reading the consumer-facing AI disclosure that NY DFS Circular Letter 2024-7 (July 11, 2024), Colorado Reg 10-1-1 (with the Oct 15, 2025 expansion and the July 1, 2026 first compliance report), California (under Prop 103 and the SB 1058-driven CDI guidance), and Connecticut Bulletin MC-25-8 require. The disclosure is not a marketing afterthought; it is a legal mechanism that supports the consumer's right to dispute, supports the DOI's enforcement authority, and supports the carrier's bad-faith defense. This lesson catalogs the consumer-disclosure requirements across the four leading states (with Nevada Bulletin 24-006 as a fifth jurisdiction watching), provides sample disclosure language for proposed-insured letters, adverse-action letters, and claim denial notices, and addresses the parallel challenge: how to preserve professional judgment - CPCU, AIC, CIC, AINS, AIAI, FCAS - when the AI workflow is doing 70% of the keystrokes. The headline rule: professional judgment is not a phrase to put on a business card; it is the substantive engagement the credentialed reviewer brings to every AI-assisted decision, documented in the file note, demonstrated in departures from AI recommendations, and tested in DOI exams and bad-faith litigation. The carrier that performs the engagement and documents it survives the audit; the carrier that rubber-stamps the AI output gets a finding letter and an extra-contractual exposure.

State-by-State Disclosure Requirements

NY DFS Circular Letter 2024-7 (July 11, 2024). NY-domiciled insurers and those writing in New York using AI/ML in underwriting and pricing must disclose to consumers in adverse-decision contexts: (a) that the decision involved AI; (b) the general factors considered; (c) the consumer's right to request human review of the decision; (d) a contact for additional explanation. The language must be clear, not buried, and provided at the time of the adverse decision. Adverse decisions include declines, non-renewals, rate increases based on AI assessment, and claim denials involving AI analysis. The Circular Letter is read alongside DFS Circular Letter No. 1 (2019) on external consumer data, the two together forming the DFS's AI/external-data framework. DFS examination authority includes targeted exams on AI use.

Colorado SB 21-169 and Reg 10-1-1. Disclosure under SB 21-169 requires the insurer to provide an explanation of adverse consumer outcomes from algorithms or predictive models. Reg 10-1-1 implements via specific language in adverse-action contexts; the consumer's right to dispute the algorithmic outcome; reference to the insurer's algorithm registry. The 2025 expansion (Oct 15, 2025) extended Reg 10-1-1 from personal-auto pricing to homeowners pricing and continues phased expansion. The first compliance report from carriers is due July 1, 2026 - meaning every Colorado-filing carrier must have AI disclosure language, an algorithm inventory, and the bias-testing exhibit submitted by that date. The Colorado Division of Insurance reviews compliance reports and may issue further guidance.

California Prop 103 plus SB 1058 expansion (2025-2026). California's existing Prop 103 framework already requires disclosure of rating factors in personal auto. SB 1058 (signaled in 2025) directed CDI to issue AI-specific disclosure guidance similar to Colorado for adverse decisions involving AI/ML. The 2026 California framework is in transition; carriers filing in California prepare AI disclosure language in advance of formal CDI guidance to avoid retroactive remediation. The California Department of Insurance market-conduct exam authority extends to AI-driven decisions.

Connecticut Bulletin MC-25-8 (2026 AI rule). Connecticut's early-2026 rule requires AI governance, fairness testing, consumer-facing AI disclosure, and DOI exam authority. Specific disclosure language is similar to NY DFS Circular Letter 2024-7. Effective for new filings 2026 forward; grandfathered for one renewal cycle on in-force policies. The Connecticut Insurance Department market-conduct division is staffed for AI examinations.

Nevada Bulletin 24-006. Issued late 2024; Nevada DOI signals expectations on AI disclosure similar to Colorado and NY. The bulletin is advisory rather than prescriptive; carriers writing in Nevada should align disclosure language with the Colorado / NY frameworks pending further rulemaking.

The eight aggressive-fairness states. Beyond the four leading states above, Washington, Massachusetts, New Jersey, and Illinois are positioned to issue similar guidance in 2026-2027. The 2026 best-practice carrier writes disclosure language that satisfies the strictest jurisdiction's requirements and applies it across the multi-state book, avoiding the operational complexity of state-by-state customization.

Sample Proposed-Insured Letter Disclosure

For new business where AI/ML influenced the rate or coverage offered:

[Date]
[Proposed Insured Name]
[Address]

Re: Quote for [Policy Type] - Quote #[N]

Dear [Name]:

We are pleased to provide the enclosed quote for [policy type]. We want you to know that the rate and coverage offered reflects our underwriting analysis, which uses automated decision systems (including AI and predictive models) in addition to underwriter review by a credentialed underwriter.

The factors considered in determining your rate include but are not limited to: [list relevant factors per state requirements - credit-based insurance score where permitted, prior claim history, vehicle characteristics for auto, driver characteristics, garaging location, property characteristics for homeowners, business operations and exposures for commercial].

If you have questions about the factors considered or our decision-making process, please contact us at [phone/email]. You may also request a review of your quote by one of our underwriters by contacting us within 30 days of this letter. We will respond to review requests within 15 business days.

Sincerely,
[Name, Title, Credentials - e.g., CPCU, AINS]

The discipline. The proposed-insured letter is the first disclosure touchpoint and sets the tone for any later adverse-action communication. Carriers that disclose at quote stage build credibility with the insured and reduce later disputes. The 30-day review window is the carrier's commitment to substantive human review; carriers that fail to respond within the window face DOI complaint exposure.

Sample Adverse-Action Letter Disclosure

For decline, non-renewal, or rate increase based on a consumer report (FCRA-compliant plus state-specific AI disclosure):

[Date]
[Insured Name]
[Address]

Re: Adverse Decision - [Policy Type] - Policy #[N]

Dear [Name]:

We have completed our review of your insurance application/policy and regret to inform you that we are unable to [decline/non-renew/issue at the requested rate]. This decision is in whole or in part based on information contained in a consumer report obtained from:

[Consumer Reporting Agency Name + Address + Phone]

The agency did not make the decision and is unable to provide you with the specific reasons for the action taken. Under the Fair Credit Reporting Act (15 U.S.C. §1681m / §615), you have the right to: (a) obtain a free copy of your consumer report from the agency within 60 days of this notice; (b) dispute the accuracy or completeness of any information in your consumer report directly with the agency.

[For NY / Colorado / Connecticut / California consumers:] This decision involved an automated decision system using AI and predictive models. The general factors considered include [list factors per state requirements]. You have the right to request a review of this decision by one of our underwriters; please contact us at [phone/email] within 30 days of this letter.

[For Colorado consumers:] Pursuant to Colorado Regulation 10-1-1, you may request additional information about the algorithm used in our decision. You may also file a complaint with the Colorado Division of Insurance at [contact].

[For NY consumers:] Pursuant to NY DFS Circular Letter 2024-7, you have the right to request additional explanation of the AI-assisted analysis that contributed to this decision. You may also file a complaint with the NY Department of Financial Services at [contact].

Sincerely,
[Name, Title, Credentials]

The FCRA overlay. The FCRA §615 adverse-action notice is independent of the state AI disclosure but must be combined operationally - one letter to the consumer that satisfies both regulatory regimes. The 60-day free-report window and the dispute right are FCRA-specific; the AI disclosure layers on top.

Sample Claim Denial Notice With AI Disclosure

For claim denial involving AI coverage analysis:

[Date]
[Insured Name]
[Address]

Re: Claim Denial - Claim #[N] - Loss Date [Date]

Dear [Name]:

We have completed our investigation of the claim referenced above. After careful review by our claims team, we regret to inform you that we are unable to provide coverage for this claim. The specific basis for our denial is: [specific exclusion or policy provision - e.g., "the loss falls within the Anti-Concurrent-Cause exclusion of your HO 00 03 policy because flood and wind both contributed to the damage and flood is an excluded peril"].

Our claim investigation included use of automated systems (including AI) in addition to review by a credentialed claims professional. The decision to deny coverage was made by [Name, Adjuster, AIC] and reviewed by [Name, Coverage Counsel, JD].

[For NY DFS / Colorado / Connecticut consumers:] If you would like to request review of our decision by additional claims personnel, please contact us at [phone/email] within 30 days. You also have the right to request additional explanation of our coverage decision.

[For California consumers receiving denial with reservation of rights:] You are entitled to independent counsel (Cumis counsel) paid by us at our expense to represent your interests with respect to coverage issues. Please contact us at [phone/email] to discuss this right and arrange engagement of Cumis counsel under California Civil Code §2860.

[For Florida consumers:] If you believe this decision is in bad faith, you may file a Civil Remedy Notice with the Florida Department of Financial Services under Florida Statute §624.155, which gives us 60 days to respond.

[For Texas consumers:] If you believe this decision violates Texas Insurance Code §541 or §542, you may file a complaint with the Texas Department of Insurance at [contact].

[Standard:] You may file a complaint with [State Department of Insurance Name + Contact]. If you are represented by counsel, we welcome contact from your counsel.

Sincerely,
[Adjuster Name, AIC]
Reviewed: [Coverage Counsel Name, JD]

The multi-jurisdiction overlay. A carrier writing in multiple states needs disclosure language that surfaces the right jurisdictional protections to the insured. The 2026 best-practice carrier maintains a state-specific overlay template that the claims-handling AI workflow populates with the consumer's state-of-residence; the credentialed reviewer verifies the overlay matches the state-of-loss and the state-of-policy.

Preserving Professional Judgment in an AI-Heavy Workflow

The credentialed reviewer in 2026 - CPCU, AIC, CIC, AINS, AIAI, FCAS, AAI, ARM, CRM - works in a workflow where the AI does 70-90% of the keystrokes. The professional-judgment risk is rubber-stamping: the reviewer clicks "approve" without substantive engagement, the file note reads "reviewed and concurred," and the audit trail does not distinguish substantive review from administrative sign-off.

Substantive engagement standard. Every AI-assisted decision requires documented substantive engagement beyond approval. Specific actions: verify the AI's analysis against source data (the loss run, the SOV, the medical records, the policy schedule); cross-reference the AI's case-law citations against Westlaw or Lexis; review the AI's policy-language interpretation against the actual policy form and edition; consider alternative interpretations the AI may have missed (the AI may have applied a standard exclusion when an endorsement modifies it); document the engagement in file notes with enough specificity that a DOI examiner reading the file two years later sees the engagement, not just the sign-off.

Departure documentation. When the credentialed reviewer departs from the AI's recommendation, document the rationale. Departures are evidence of professional judgment in action. An artifact with zero departures across all AI-assisted decisions is a rubber-stamp signal that DOI examiners and bad-faith plaintiffs may exploit. The 2026 best-practice carrier tracks reviewer-level departure rates and surfaces reviewers with unusually low departure rates for management review.

Training cadence. Annual training on AI workflow, departure-documentation discipline, emerging case law (the Texas Stowers analyses, the Florida CRN landscape, the California Cumis triggers), regulatory updates (the Colorado Reg 10-1-1 compliance report cycle, the NY DFS Circular Letter 2024-7 exam findings), and new AI failure modes (hallucinated case citations, phantom endorsements, fabricated IRIS ratios per the catch-hallucinations lesson). Documented in the MRM registry. The DOI examiner asks for training records during financial-condition examinations.

Professional credential maintenance. CPCU, AIC, CIC, AINS, AIAI, FCAS each require continuing education. The 2026 best-practice carrier ensures continuing-education includes AI-workflow proficiency, AI-output review discipline, and ethics-of-AI-use coursework. Documented in the credentialing body's records (CPCU through The Institutes, FCAS through CAS, AIAI through The Institutes' new AI in Insurance designation) plus the carrier's MRM registry. The AM Best 2026 readiness survey asks about credential maintenance for AI-touching roles.

Workflow design. The AI workflow itself preserves judgment when designed to require explicit reviewer engagement at decision points. Sign-off requires specific reviewer actions, not just an "approve" button - for example, the reviewer must enter a 3-sentence rationale, must check specific verification boxes (case citations verified, form editions confirmed, policy-language reconciled), and must affirm departure-or-concurrence with the AI's recommendation. System prompts explicitly state "do not approve without review of [specific items]." Workflow logs every reviewer action - time spent reviewing, items examined, departures.

The Rubber-Stamp Risk and Its Mitigation

Rubber-stamp pattern: the credentialed reviewer signs every AI output without substantive engagement; the same time is spent per review regardless of complexity; no departures from AI recommendations; file notes are consistent boilerplate ("reviewed and concurred"). The pattern is detected via workflow analytics - reviewer-action statistics; departures-per-N-reviews ratio; review-time-per-decision distribution; review-time variance across complexity (a reviewer who spends the same 90 seconds on a $25K straightforward subro file and a $4M multi-claimant BI file is a candidate for rubber-stamp review).

Bad-faith litigation discovery surfaces these patterns - plaintiff's counsel obtains the workflow analytics in discovery and demonstrates rubber-stamping to the jury. DOI market-conduct exam findings highlight rubber-stamp patterns as systemic risk. The carrier's AM Best 2026 readiness survey response on AI governance is undermined by rubber-stamp signals.

Mitigation. Workflow design requires specific reviewer actions (per the workflow-design point above). Training reinforces the engagement standard with concrete examples and bad-faith case studies. Quarterly reviewer-performance review includes departure rates, engagement signals, and time-variance analysis. HR and management consequences for sustained rubber-stamp patterns - coaching, retraining, role change. The 2026 best-practice carrier monitors rubber-stamp signals on every credentialed reviewer; carriers that do not monitor see the pattern emerge in bad-faith litigation when it is too late to remediate.

The Cross-Jurisdiction Discipline and the National Carrier

A national carrier with consumers in all 50 states and policies subject to multiple state regulatory frameworks must operationalize the disclosure discipline across the book. The 2026 best-practice pattern: write disclosure language that satisfies the strictest jurisdiction (currently Colorado plus NY DFS); apply uniformly across the book; let state-specific overlays surface jurisdiction-specific protections (Florida CRN, Texas §541/§542, California Cumis/Brandt) at the point of decision.

The cross-jurisdiction discipline reduces operational complexity (one disclosure standard rather than 50) and reduces audit exposure (no jurisdiction is under-served). The carrier's general counsel signs off on the unified disclosure language; the chief compliance officer monitors disclosure-delivery metrics; the AI governance committee reviews quarterly. The Boston manufacturing-renewal stewardship example (the $1.2M premium / 22% rate hike / 45-day window) is the type of multi-state-policy scenario where the disclosure overlay matters - the manufacturing operations span Massachusetts, Connecticut, and Rhode Island, and the AI-driven rate analysis surfaces different disclosures depending on which state DOI has jurisdiction over the renewal.

The 2027 horizon. Additional state DOIs are expected to issue AI disclosure guidance in 2027 - Washington, Massachusetts, New Jersey, Illinois are queued. Federal activity at OCC, FRB, FTC, and HHS OCR may layer additional disclosure requirements for AI in insurance-adjacent financial services. The Mental Health Parity NQTL Tri-Agency 2024 final rule adds parallel disclosure-style requirements for L&H carriers. The carrier that has built the disciplined disclosure framework in 2026 absorbs 2027 regulatory changes without operational disruption; the carrier that has not built the framework faces a remediation cycle that absorbs compliance bandwidth at exactly the wrong moment.

The Regulator's View and the AM Best Analyst's View

The DOI market-conduct examiner approaches the disclosure-and-professional-judgment discipline through a structured framework: sample 25-40 adverse-decision files; verify that the disclosure language was delivered; check that the disclosure matches the state's requirement; review the file note for substantive engagement signals; cross-check the reviewer's credentials against the carrier's MRM registry; ask for departure statistics across the reviewer's queue; request training records for the reviewer in question. The examiner's finding letter cites specific files, specific disclosure failures, and specific engagement gaps. Remediation requires file-by-file correction plus systemic process change documented to the DOI within an agreed window.

The AM Best analyst's view. AM Best's 2026 AI-readiness framework explicitly includes consumer-disclosure governance and professional-judgment preservation as survey-not-rating elements. The analyst asks the carrier's senior management whether disclosure language is uniform across the multi-state book, whether the carrier monitors rubber-stamp signals on credentialed reviewers, whether training cadence is current, and whether the AM Best 2026 framework's specific questions on AI governance can be answered with documented evidence. The framework is survey rather than rating methodology, but a carrier that cannot answer credibly may face elevated scrutiny on future rating cycles - and treaty brokers reading the AM Best file form similar judgments about cedent quality.

The treaty broker's view. The cedent's reinsurance accounting director and the treaty broker (Aon, Guy Carpenter, Howden Tiger, McGill, BMS) read the carrier's AI-governance posture before placing renewals. A cedent with clean disclosure language, documented training, and a defensible professional-judgment standard places renewals at favorable terms; a cedent with disclosure gaps and rubber-stamp signals faces tighter AI representations from the reinsurer (per the treaty wording markup lesson's 2026 AI-driven UW representations discussion) and may see rate, capacity, or sunset-clause concessions extracted as a result.

Key Takeaways

• Four leading states plus one watching: NY DFS Circular Letter 2024-7 (July 11, 2024), Colorado SB 21-169 / Reg 10-1-1 (Oct 15, 2025 expansion, July 1, 2026 first compliance report), California Prop 103 + SB 1058 (2025 signal), Connecticut Bulletin MC-25-8 (2026), Nevada Bulletin 24-006 (advisory).
• NY DFS Circular Letter 2024-7: adverse-decision contexts require (a) AI involvement disclosed; (b) general factors; (c) right to human review; (d) contact for explanation. Clear, not buried, at time of decision. Read alongside Circular Letter No. 1 (2019) on external consumer data.
• Colorado Reg 10-1-1: explanation of adverse outcomes from algorithms; consumer right to dispute; reference to algorithm registry. Compliance report due July 1, 2026; expansion from auto to homeowners Oct 15, 2025; phased expansion continues to L&H and commercial.
• Proposed-insured letter sample includes factors-considered language and right-to-underwriter-review within 30 days. Carrier responds to review requests within 15 business days. Disclosure at quote stage reduces later disputes.
• Adverse-action letter sample combines FCRA §615 requirements (consumer report agency contact, free report right within 60 days, dispute right) with state-specific AI disclosure overlays for NY, Colorado, Connecticut, California consumers.
• Claim denial notice sample includes denial basis, AI involvement disclosure, reviewer credentials, and state-specific overlays. NY review request, California Cumis offer under Civil Code §2860, Florida CRN notice under §624.155, Texas §541/§542 complaint information, standard DOI complaint right.
• Professional-judgment preservation requires substantive engagement standard, departure documentation, training cadence, credential maintenance, workflow design with explicit reviewer-action requirements. Documented in MRM registry; tested in DOI exams and bad-faith litigation.
• Rubber-stamp risk: same review time across complexity, zero departures, boilerplate file notes. Detected via workflow analytics (departures-per-N-reviews ratio, review-time variance); surfaced in bad-faith litigation discovery; mitigated by workflow design plus training plus quarterly reviewer-performance review plus HR consequences.
• Continuing education for credentialed reviewers in 2026 includes AI-workflow proficiency, AI-output review discipline, and AI ethics coursework. CPCU through The Institutes; FCAS through CAS; AIAI through The Institutes' AI in Insurance designation. Documented in credentialing-body and carrier MRM records. AM Best 2026 readiness survey asks about credential maintenance for AI-touching roles.