A claim file note that survives a bad-faith suit reads like a credentialed reviewer's signed memo, not like an AI transcript. A SERFF filing memorandum that survives a Colorado Division of Insurance Reg 10-1-1 exam reads like an FCAS's defensible decision narrative, not like an AI draft. A bordereau report that survives an MGA audit by the fronting carrier's delegated-authority team reads like a compliance officer's signed certification with reconciliation evidence, not like an AI table dump. This lesson catalogs the file-note documentation patterns for six high-stakes AI-assisted decisions - UW decline, claim denial, reservation of rights, rescission, SERFF filing, bordereau reporting - and provides sample file-note language for each that satisfies the four-element discipline: (a) the trigger that surfaced the decision, (b) the AI's analysis including invocation ID and model version, (c) the credentialed reviewer's substantive engagement and any departures from the AI's recommendation, (d) the signature with credentials. The discovery-cooperation framework that follows is what turns documented AI-assistance from a liability into an asset under Texas Insurance Code §541, Florida §624.155, California Insurance Code §790.03, the NAIC Unfair Claims Settlement Practices Model Act §4, and the NAIC Model Bulletin §4.1–§4.4. The headline rule: the file note is the artifact that survives discovery; if the file note says "system generated" or "AI determined" without credentialed-reviewer engagement, the carrier has built its own bad-faith exposure and a discoverable admission against interest.

Six High-Stakes AI-Assisted Decisions

The six decisions below are the highest-frequency points where AI-assisted output becomes a discoverable artifact in litigation, an examined artifact in a DOI market-conduct exam, or an audited artifact in an MGA capacity-provider review. Each requires a documentation pattern; each pattern shares the same four-element skeleton but loads different content per decision type.

Decision 1 - UW decline. AI surfaces declination signals via Cytora's appetite-scoring engine or Convr's Risk 360 scorecard or Federato RiskOps' triage rules. Credentialed UW (AINS, CPCU, or licensed P&C producer) decides. Documentation captures both AI input and UW judgment; FCRA §615 pre-notice / adverse-action analysis attaches if the decline relied on consumer-report data.

Decision 2 - Claim denial. AI analyzes coverage via Five Sigma's coverage analysis or Guidewire ClaimCenter's coverage module or Roots Automation's cognitive intake worker. Adjuster (AIC, SCLA) and coverage counsel decide. Documentation captures coverage analysis, the form-edition validation (CG 00 01, HO 00 03, CA 00 01 with edition dates), case-law references, reviewer engagement, denial rationale.

Decision 3 - Reservation of rights. AI surfaces RoR triggers (late notice, intentional act, business-use exclusion, policy-period question). Coverage counsel decides. Documentation captures specific reservation grounds, the Cumis offer in California, the ongoing-defense framework, and the final coverage disposition upon resolution.

Decision 4 - Rescission. AI surfaces material-misrepresentation signals via cross-reference against LexisNexis Risk Classifier, the MIB underwriting code disclosure, MVR, CLUE, Milliman IntelliScript, or ExamOne ScriptCheck. Coverage counsel decides. Documentation captures application review, misrepresentation materiality analysis, statutory-window compliance, notice-of-rescission with contest window.

Decision 5 - SERFF filing. AI drafts the rate filing memorandum and the actuarial certification via Akur8's filing-generation module. Pricing actuary (FCAS, MAAA) and chief actuary review and sign per ASOPs 23/41/56. Documentation captures AI invocation log, actuarial customization scope, ASOP attestations, bias-testing exhibit (Colorado Reg 10-1-1, NY DFS Circular Letter 2024-7), sign-off chain.

Decision 6 - Bordereau reporting. AI generates the bordereau from VIPR or Brisc or the MGA's internal PAS. MGA compliance officer (AIAI, AIC, AAI) reviews and certifies; the fronting carrier's delegated-authority team audits monthly or quarterly. Documentation captures data extraction, reconciliation against the underwriting system, cession entries against treaty terms, surplus-lines stamping-office verification, compliance certification.

File-Note Pattern - UW Decline

The UW decline file note must accomplish four things at once: identify AI involvement explicitly with invocation ID; identify the credentialed reviewer's substantive engagement; specify the basis and statutory or appetite-rule authority for the decline; flag FCRA applicability and document the pre-notice and adverse-action notice if a consumer report drove the decision. The sample below is the template a Federato or Cytora deployment can pre-populate, with the underwriter customizing the underlined sections.

UW DECLINE - Account: [Named Insured]. Submission Date: [Date]. Reviewing Underwriter: [Name, AINS/CPCU]. Decision: Decline. Effective: [Date].

Basis: Account does not meet appetite criteria. Specific factors:
(a) Loss ratio prior 3 years: 142% (appetite threshold 75%);
(b) Class code 5403 (interior carpentry) inside appetite but with adverse modifier: open OSHA inspection 2024 with unresolved citations;
(c) Geographic exposure: 18% of payroll concentrated in three high-severity zip codes flagged by Verisk territory model v2026-Q1;
(d) Prior-carrier non-renewal for cause (2024 cycle), confirmed via ISO ClaimSearch cross-reference.

AI clearance system [Cytora v.4.8.2 / invocation ID 2026-04-12-X4582 / appetite guide v2026-04 PDF, section 3.2 warehousing] surfaced these factors via appetite-scoring engine; score 23/100 (below 45 threshold for decline routing). UW review confirmed accuracy of underlying data via Applied Epic policy record + LexisNexis loss-run reconciliation + ISO ClaimSearch query 2026-04-12 13:47. UW concurred with decline recommendation; no override exercised.

Producer notified [Date] via standard decline letter referencing appetite-guide section 3.2 line items without disclosing competitive intelligence or proprietary scoring. FCRA notice required: NO - decline not based on consumer report; Cytora's appetite score derived from policy-application data and ISO ClaimSearch carrier-contributed loss history, neither of which qualifies as a consumer report under FCRA §603(d). Documented per NAIC Model Bulletin §4.4.

Signed: [Name, AINS, CPCU], [Date]. Reason chain attached as JSON export from Cytora.

What this note achieves. Identifies AI involvement explicitly with the invocation ID and model version so a §4 examiner can re-create the run. Identifies the credentialed reviewer's substantive engagement - not "reviewed and concurred" but "confirmed accuracy of underlying data via [three specific verification sources]." Specifies the basis and authority for decline against the appetite guide line items rather than subjective judgment. Flags FCRA non-applicability with the statutory citation, anticipating the examiner question rather than waiting for it. Provides a discoverable audit trail that a plaintiff's E&O counsel cannot characterize as black-box decisioning.

Common failure modes on the decline note. Saying "Cytora determined the account is outside appetite" without the appetite-guide line item reference - the §4.4 reason chain is incomplete. Saying "credit-based scoring drove the decline" when a credit factor was a feature but not the sole driver - that mischaracterizes the FCRA trigger and forces an adverse-action notice that may not be required. Failing to capture the producer-notification language verbatim - discoverable inconsistency between the file note and the producer letter is a common impeachment vector.

File-Note Pattern - Claim Denial

The claim denial file note carries the highest bad-faith stakes of any artifact in the claims pipeline. Texas §541 punitive damages, Florida §624.155 Civil Remedy Notice (CRN) with a 60-day cure window, California §790.03 fair-claims unfair practices - each creates a specific liability exposure when a denial is later proven to have lacked good-faith investigation. The file note must show the coverage analysis, the AI's contribution, the adjuster's verification, the coverage counsel's engagement, and the specific exclusion or coverage condition that authorized the denial.

CLAIM DENIAL - Claim Number: [N]. Insured: [Named Insured]. Date of Loss: [Date]. Adjuster: [Name, AIC]. Coverage Counsel: [Name, JD].

Loss: Property damage from water leak. Insured reports loss caused by burst pipe; investigation revealed leak originated from sewer-backup at municipal connection.

Coverage Analysis: ISO HO 00 03 (10 11) policy. Investigation confirmed sewer-backup origin per municipal sewer-camera report dated [Date] (attached) and plumber inspection report dated [Date] (attached). Policy excludes sewer-backup damage (Section I - Exclusion 5.a). Anti-Concurrent-Cause language in HO 00 03 lead-in to Section I Exclusions confirms exclusion applies even where covered cause concurs. AI coverage analysis [Five Sigma invocation ID 2026-03-22-Y8214, model version v2026-Q1] retrieved Sections I-A, I-Exclusion-5, Florida case law on sewer-backup exclusion enforceability. Cited Sun Insurance v. Florida (2018) confirming exclusion enforced in Florida; cited Mierzwa v. Florida Windstorm Underwriting Ass'n (2003) on ACC clause enforcement.

Adjuster's review: confirmed AI coverage analysis. Verified specific facts via plumber inspection report (attached) and municipal sewer report (attached). Confirmed HO 00 03 (10 11) edition matches policy declarations. Confirmed Exclusion 5.a present in attached policy. Coverage counsel reviewed and concurred; no ambiguity argument on Section I 5.a found.

Decision: Deny claim. Specific exclusion cited (Section I, Exclusion 5.a sewer-backup, with ACC lead-in clause). Florida bad-faith framework considered (§624.155); denial supported by clear exclusion, two independent factual investigations, and on-point case law. Denial letter dated [Date] sent via certified mail with copy of relevant policy pages.

Florida CRN trigger consideration: insured may file CRN within statutory window; cure-window monitoring established in ClaimCenter task queue (60-day SLA). If CRN received, immediate escalation to claims VP + Florida coverage counsel + outside coverage counsel for response within cure window.

Signed: [Adjuster Name, AIC], [Date]. Reviewed: [Coverage Counsel Name, JD], [Date]. Five Sigma invocation JSON attached.

The two-source factual rule. A claim denial relying on a single fact source (one inspection report, one recorded statement, one site photo) is more vulnerable to bad-faith attack than one supported by two or more independent sources. The Five Sigma analysis flagged the exclusion; the plumber report and municipal sewer report independently confirmed the factual predicate. The file note captures both. In a Texas §541 trial, the carrier's expert can testify to the two-source investigation and the AI's confirmatory role; in a Florida §624.155 CRN response, the carrier's cure-window letter can attach the two reports and the AI analysis as good-faith evidence.

File-Note Pattern - Reservation of Rights (California)

The California RoR file note carries the Cumis dimension. San Diego Federal Credit Union v. Cumis Insurance Society (1984) established that a coverage conflict between insurer and insured creates the insured's right to independent counsel at the insurer's expense, codified at California Civil Code §2860. Failing to advise the insured of Cumis rights when the conflict exists is a regulator-attended finding and a bad-faith multiplier. The file note must capture the RoR grounds, the AI's flag, the coverage counsel's engagement, and the Cumis advisement with counsel selection.

RESERVATION OF RIGHTS - Claim Number: [N]. Insured: [Named Insured]. Date of Loss: [Date]. Coverage Counsel: [Name, JD].

Claim: Bodily injury action against insured arising from premises liability matter. Investigation reveals potential trigger of intentional-act exclusion based on prior-incident pattern (3 similar incidents in 18 months per insured's own incident log).

Coverage Analysis: AI analysis [Five Sigma invocation ID 2026-02-15-Z3147, model version v2026-Q1] flagged potential intentional-act exclusion via retrieved policy CGL CG 00 01 (04 13) Section I Coverage A 2.a Exclusions - Expected or Intended Injury. Coverage counsel reviewed; concurred on potential application pending factual development through underlying litigation discovery.

Decision: Reserve rights on intentional-act exclusion while defending claim. RoR letter drafted [Date]; sent via certified mail to insured; copy to defense counsel of record.

California Cumis trigger: Reservation of rights creates conflict of interest per Cumis v. San Diego Federal CU (1984) codified at Civil Code §2860. Insured advised in writing [Date] of right to independent Cumis counsel paid by carrier at rate per Civil Code §2860 (rate cap formula referenced). Insured selected: [Cumis Counsel Name, JD, Firm] effective [Date]. Cumis counsel engaged [Date]; coordination with appointed defense counsel established via written protocol; billing review by carrier monthly.

Ongoing: monitor defense; periodic coverage review at each material litigation event (motion practice, deposition of key witnesses, mediation); final coverage determination upon resolution of underlying matter. RoR letter revisited if facts develop confirming or rebutting intentional-act trigger.

Signed: [Coverage Counsel, JD], [Date]. AI Invocation Logs Attached. Cumis Advisement Letter Attached.

The Cumis discipline. California Department of Insurance market-conduct exams routinely review RoR files for Cumis advisement. A file note that says "RoR issued; insured advised of Cumis rights" without the date, the rate citation, the counsel-selection record, and the coordination protocol is incomplete. The carrier's outside coverage counsel and inside claims management both review the file note for completeness before the RoR letter ships. AI can draft the Cumis advisement language from a template, but the counsel-selection step is human-only and the protocol must be paper.

File-Note Pattern - Rescission

Rescission undoes the policy ab initio - as if it never existed. The remedy is powerful but constrained by state-specific statutes that require materiality, reliance, and a rescission window (typically 12-24 months from inception). Failing any element exposes the carrier to bad-faith for wrongful rescission. AI cross-references the application against external data sources to surface misrepresentation candidates; coverage counsel decides.

RESCISSION - Policy Number: [N]. Named Insured: [Name]. Original Effective Date: [Date]. Coverage Counsel: [Name, JD].

Basis: Material misrepresentation on application. Application question 14 (prior claims past 5 years) answered NO; CLUE history shows 3 prior claims at prior addresses 2021-2023; ISO ClaimSearch cross-reference confirms; LexisNexis Risk Classifier confirms.

AI Application Review: [Five Sigma invocation ID 2026-01-08-W9742, model version v2026-Q1] cross-referenced application against CLUE + MVR + LexisNexis Risk Classifier + MIB (where applicable for L&H). Discrepancy identified on question 14. AI surfaced for coverage counsel review; AI's recommendation: investigate materiality; AI did not recommend immediate rescission.

Coverage Counsel Review: confirmed AI cross-reference accuracy via direct CLUE and ISO ClaimSearch queries dated [Date]. Reviewed application question 14 wording; clear; not ambiguous. Application signed by named insured; material to underwriting decision (would have declined or rated differently with accurate disclosure - UW manager [Name, CPCU] affirmed in writing).

Statutory Compliance: [State, Statute §] permits rescission for material misrepresentation if (a) misrepresentation was material (affirmed by UW manager), (b) carrier relied on application in underwriting decision (production binding file demonstrates reliance), (c) within statutory rescission window (12 months from inception in this state; current date within window).

Decision: Rescind policy ab initio. Premium refund issued in full. Notice of rescission sent via certified mail [Date]. Rescission letter explains basis, cites specific application question and contradictory data sources, and provides 30-day window for insured to contest with documentation.

Signed: [Coverage Counsel, JD], [Date]. Five Sigma JSON, CLUE printout, ISO ClaimSearch query result attached.

The materiality affidavit. The UW manager's written affirmation that the misrepresentation was material to the underwriting decision is the load-bearing element. Without it, the carrier's rescission claim is vulnerable to a plaintiff's argument that the misrepresentation was peripheral. AI can draft the affidavit template; the UW manager signs after reviewing the underwriting file and confirming that the disclosed information would have produced a different decision (decline, rate adjustment, exclusion endorsement).

File-Note Pattern - SERFF Filing

The SERFF filing file note documents the AI's role in drafting the rate filing memorandum, the pricing actuary's customization, the ASOP attestations, the bias-testing exhibit, and the sign-off chain. The Colorado Division of Insurance and the NY DFS read this file note when they audit the filing; the AISET Exhibits A/B/C/D pull from it; the AM Best analyst's annual readiness survey samples from it.

SERFF FILING - State: Colorado. Filing Number: [SERFF Tracking ID]. Effective: [Proposed Date]. Filing Manager: [Name]. Pricing Actuary: [Name, FCAS, MAAA]. Chief Actuary: [Name, FCAS, MAAA].

Filing: Personal-auto BI severity rate change +5.8% statewide based on 2023-2025 experience. Companion filings: California (Prop 103 prior-approval) and New York (DFS Circular Letter 2024-7 disclosure requirements).

AI Involvement: Akur8 filing generation [invocation ID 2026-05-14-A2847, model version Akur8 Deploy v2026-Q2] produced 12-page memorandum draft from the candidate model card, the rate-change indication exhibit, and the state-specific constraint set (Colorado Reg 10-1-1, AISET Exhibit A bias-testing prompts). Pricing actuary customized 6-10 hours adding company-specific narrative (competitive positioning vs. Akur8 Discover peer median +5.4%), prior-filing reconciliation (last filed +3.2% effective 2024-Q3), Colorado-specific Reg 10-1-1 disclosures (algorithm inventory entry reference, July 1 2026 compliance report alignment).

Bias Testing: BISG-imputed disparate-impact ratios 0.91 / 0.94 / 0.97 for the three protected-class proxies tested, all above 0.80 four-fifths threshold and above Reg 10-1-1's 0.85 expectation. Geographic-proxy test on 22 territory codes per NY DFS proxy-test methodology - 3 dropped at feature engineering for correlation > 0.30 with race-imputed BISG indicator. SHAP / PDP / ALE artifacts attached as Exhibit B. Sensitivity analyses on credit-factor magnitude included as Exhibit C.

ASOP Compliance: ASOP 23 data-quality memo (signed by data steward), ASOP 41 communications attestation signed by pricing actuary, ASOP 56 modeling attestation signed by external peer reviewer (firm: [Name], FCAS, MAAA, independent of company).

Filing Package: rate filing memorandum (12 pages) + actuarial certification (1 page) + rate-change indication exhibit + bias-testing exhibit (16 pages) + model card by reference (URL + commit hash) + response-to-objection template (pre-drafted 8 common Colorado DOI objection patterns from Matrisk database).

Sign-off Chain: pricing actuary [Date], chief actuary [Date], general counsel review [Date], filing manager submission via SERFF portal [Date]. Multi-state coordinator confirmed CA and NY companion filings same-day.

Signed: [Chief Actuary, FCAS, MAAA], [Date].

The Akur8 invocation log as discoverable artifact. Colorado Reg 10-1-1's algorithm inventory entry requires the carrier to maintain a record of when and how the model was used. The Akur8 invocation ID and model version tie the filing memorandum to a specific deployment run; the bias-testing exhibit ties the deployment to a specific in-force book snapshot; the ASOP 56 modeling attestation ties the deployment to a peer-reviewed methodology. The three together create the defensible chain that a Colorado examiner reads in 2026.

File-Note Pattern - Bordereau Reporting

The MGA bordereau file note is the compliance officer's certification that the bordereau accurately reflects bound policies, that delegated authority limits were not exceeded, that surplus-lines stamping was completed where applicable, and that aggregate exposures fall within treaty caps. The fronting carrier's delegated-authority team audits monthly or quarterly; the audit posture in 2026 has tightened materially as fronting arrangements expanded and Lloyd's Bermuda and London market syndicates pulled back from automatic delegation.

BORDEREAU REPORT - Month: [Month/Year]. MGA: [Name]. Capacity Provider: [Name]. Compliance Officer: [Name, AIAI, AIC].

Bordereau Contents: 47 new bound policies + 12 endorsements + 3 cancellations + 4 claims (FNOLs reported during the month).

AI Bordereau Generation: [VIPR delegated authority platform invocation ID 2026-04-30-B5821, generation script v2026-Q2] extracted policy data from underwriting system; applied delegated authority validation rules from the MGA agreement Schedule A; generated bordereau table with all required fields per capacity provider template (named insured, FEIN, effective dates, premium, limits, retention, cession percentage, surplus-lines tax remitted, broker of record).

Compliance Review: verified bordereau completeness (47/47 policies, 12/12 endorsements, 3/3 cancellations, 4/4 claims); cross-referenced to underwriting system 100% (no orphan records, no PAS-only records missing from bordereau). Authority compliance: all bound within delegated per-risk and aggregate limits (one policy approached $4.8M property cap of $5M - within authority but flagged for capacity-provider attention). Surplus-lines verification: 14 SL policies; diligent-effort affidavits documented for FL (8), TX (4), CA (2); stamping-office filings completed within state SLA (FSLSO 30-day window, SLTX 60-day, SLA-CA monthly batch). Exposure aggregation review: FL Tier-1 wind $32M / $40M cap (80%); TX Gulf $14M / $20M (70%); CA EQ $8M / $15M (53%) - all within treaty per-peril aggregate.

Compliance Issues: SB2 (sub-broker) placed one $4.2M property approaching authority - coached and referred to MGA prior to bind; no breach. One ELANY 60-day filing window approaching expiration - expedited by ops team; filed Day 56 within window.

Certification: I, [Name, AIAI, AIC], certify the above bordereau is complete and accurate per delegated authority and surplus-lines compliance requirements as of [Date]. Cession entries reconcile to treaty terms per Schedule B of the MGA agreement.

Signed: [Compliance Officer, AIAI, AIC], [Date]. Distribution: Capacity Provider [Name, Delegated Authority Team], MGA Principal.

The reconciliation evidence. The fronting carrier's audit examines the bordereau against the MGA's PAS, against the surplus-lines stamping office filings, and against the treaty exposure aggregates. The compliance officer's signed certification carries weight only if the underlying reconciliation evidence is preserved. VIPR or Brisc's audit log captures the reconciliation queries; the MGA preserves the log for the treaty audit window (typically 7 years post-treaty termination).

Discovery-Cooperation Framework

In bad-faith litigation, in a NAIC market-conduct exam, in a DOI Reg 10-1-1 algorithm-inventory request, in an AISET Exhibit C audit, in an MGA delegated-authority audit, in a Lloyd's coverholder review, in an AM Best readiness survey - the carrier may be asked to produce AI workflow documentation. The discovery-cooperation framework is what turns AI-assistance from a liability into an asset.

The seven-element preservation discipline. (1) AI invocation logs preserved with retention policy aligned to claim file retention (typically 7-10 years for P&C; longer for L&H per state) - the invocation ID, model version, prompt, retrieval-augmented-generation corpus version, and output captured at each call. (2) Verification-layer logs preserved similarly - form-edition checks, named-insured cross-checks, sublimit reasonableness, loss-run reconciliation, treaty-constraint enforcement results. (3) Escalation register preserved - every override, every supervisor approval, every coverage-counsel referral. (4) Prompt library versioning preserved - system prompts at the role level (underwriter, adjuster, producer, actuary) with commit history. (5) RAG corpus version history preserved - which version of the forms library, the appetite guide, the treaty wording, the SERFF objection database the AI consulted. (6) MRM committee minutes preserved - model validation reports, peer reviews, bias-testing summaries, drift-monitoring outputs. (7) Carrier's documented AI policies preserved - the NAIC §4.2 governance document, the AI use policy, the consumer-disclosure standard.

The discovery posture in 2026. Plaintiff's E&O counsel and bad-faith counsel routinely request AI-assistance documentation. The carrier that produces a clean preservation package - invocation IDs, versioned prompts, MRM minutes, verification logs - defeats the "black-box decisioning" narrative immediately. The carrier that produces incomplete or inconsistent documentation invites a sanctions motion and a punitive-damages instruction at trial. The 2026 best-practice carrier maintains AI workflow as a fully discoverable system and designs its documentation discipline accordingly.

The MGA audit posture. Fronting carriers in 2026 increasingly demand the same preservation discipline from delegated-authority MGAs. The MGA agreement Schedule C now commonly includes the seven-element preservation list as a condition of authority retention. MGAs that cannot produce a clean audit package lose authority at the next renewal cycle; capacity providers withdraw paper.

Key Takeaways

• Six high-stakes AI-assisted decisions require disciplined documentation: UW decline, claim denial, reservation of rights, rescission, SERFF filing, bordereau reporting. Each shares the four-element skeleton (trigger + AI analysis with invocation ID + credentialed reviewer engagement + signature with credentials) but loads decision-specific content.
• File-note discipline: never "system generated" or "AI determined" alone. That language builds bad-faith exposure under Texas §541, Florida §624.155, California §790.03, and the NAIC Unfair Claims Settlement Practices Model Act §4. The credentialed reviewer's substantive engagement must be documented with specific verification steps, not "reviewed and concurred."
• Identify AI involvement explicitly: invocation ID, model version, RAG corpus version, prompt reference. Discoverable; defensible; supports NAIC market-conduct exam, Colorado Reg 10-1-1 algorithm inventory, NY DFS Circular Letter 2024-7 proxy test, AISET Exhibits A/B/C/D, AM Best readiness survey, and any E&O or bad-faith defense.
• UW decline file note covers basis against appetite-guide line items, AI clearance (Cytora or Convr) with invocation ID, UW verification via independent sources, FCRA §615 applicability analysis with statutory citation, signature with credentials. The two-source factual rule applies whenever a consumer report may have been a feature.
• Claim denial file note covers Five Sigma coverage analysis with form-edition validation (CG 00 01, HO 00 03, CA 00 01 with edition dates), Anti-Concurrent-Cause clause where applicable, two independent factual sources, adjuster review, coverage counsel concurrence, Florida CRN trigger with cure-window monitoring, certified-mail delivery.
• California RoR file note covers reservation grounds, Five Sigma AI flag, coverage counsel review, Cumis advisement under Civil Code §2860 with rate citation, counsel selection date, coordination protocol with appointed defense counsel. CDI market-conduct exams review RoR files for Cumis completeness.
• Rescission file note covers misrepresentation basis, AI cross-reference against CLUE + ISO ClaimSearch + LexisNexis Risk Classifier + MIB where applicable, materiality affidavit from UW manager, statutory-window compliance, notice of rescission with 30-day contest window.
• SERFF filing note covers Akur8 invocation log with model version, pricing-actuary customization scope, bias-testing exhibit per Colorado Reg 10-1-1 and NY DFS Letter 2024-7, ASOP 23/41/56 attestations, sign-off chain through chief actuary and general counsel. The Akur8 invocation ID ties the filing to a specific deployment run for algorithm-inventory traceability.
• Bordereau note covers VIPR or Brisc AI generation with invocation ID, compliance review (completeness + authority + surplus-lines + treaty-aggregate), compliance-issue log, AIAI/AIC certification. Fronting carriers in 2026 increasingly demand the seven-element preservation discipline from MGAs as a condition of delegated-authority retention.
• Discovery-cooperation framework preserves AI workflow as fully discoverable system: invocation logs + verification logs + escalation register + prompt library versions + RAG corpus history + MRM minutes + AI policies. Best-practice 2026 carrier designs documentation as if it will all be discovered, because under NAIC §4, Colorado Reg 10-1-1, NY DFS 2024-7, and AISET, it will be.