The file note is the durable record of insurance work - the entry in the policy administration system or claim system or AMS that survives the underwriter / adjuster / producer / actuary who created it, the carrier or agency that holds it, and the years between creation and the eventual examination, litigation, rating-agency review, or discovery production that surfaces it. Documenting AI involvement in the file note is no longer optional in 2026 - NAIC Model Bulletin §4 expects governance documentation of AI use; Colorado Reg 10-1-1 requires algorithm-inventory traceability; NY DFS Circular Letter 2024-7 expects governance discipline; FCRA, ERISA, MHPAEA enforcement frameworks expect specific documentation of methodology; discovery in bad-faith litigation increasingly demands AI-prompt-log production. This lesson walks the file-note standards by persona - underwriter file notes in PolicyCenter or Duck Creek or Sapiens or Majesco; adjuster file notes in ClaimCenter or claim platform; producer activity logs in Applied Epic or AMS360 or Vertafore; actuarial files in reserving software or rate-filing system - plus where to store the prompt log, how to handle a discovery request that includes AI-generated content, and how to preserve work product and attorney-client privilege when AI is involved in a defense-panel claim. The lesson closes with discovery-cooperation framework for a bad-faith suit and sample file-note language for each artifact tier.

The File-Note Discipline

The file note documents the human's work on the case - what was done, when, why, and with what tools or evidence. AI involvement is part of the work and belongs in the note. The 2026 mature carrier or agency operates a structured file-note standard that includes specific elements for AI-touched artifacts: (1) AI assistance acknowledgment - explicit statement that AI assisted in producing the artifact; (2) Prompt version reference - system prompt or template version used; (3) Model version reference - specific AI model (e.g., claude-3.5-sonnet-2026-05); (4) Verification reference - checklist or verification process applied; (5) Human reviewer identification with sign-date; (6) Cross-functional sign-off references where applicable; (7) Prompt-log archive location.

The discipline serves three purposes: NAIC §4 personal-review attestation; chain-of-custody for examination and litigation; reproducibility for future review or audit. Without the discipline, AI involvement disappears into routine work product and emerges unexpectedly during discovery or examination. With the discipline, AI involvement is documented, traceable, and defensible. The cost is minimal (1-2 minutes of file-note discipline per AI-touched artifact); the value is significant (avoided exposure in examination and litigation).

Underwriter File-Note Pattern

The underwriter operates in Guidewire PolicyCenter, Duck Creek, Sapiens, or Majesco; file notes are typed into the underwriting record. Sample file-note language for an AI-assisted submission triage:

"Submission triage AI-assisted via prompt underwriter_v3.2 (Claude 3.5 Sonnet 2026-05) on submission [number]. Source documents: ACORD 125 [date], ACORD 140 [date], 5-year loss run [date], SOV [date]. AI generated structured submission summary with classification (NAICS 332710, SIC 3599, ISO GL Class Code 51315), exposure analysis ($48M revenue, 95 employees, three locations OH/IN/KY), loss-history summary (clean except 2022 GL claim $47K customer-property condition), appetite reference (Hartford GI-2026.M.001 §3.2 matched preferred tier), and treaty considerations (within standard single-risk limits). Verified per submission-summary checklist v2.1 by [underwriter name] [date]; no defects identified. AI-assisted summary attached to file. Prompt-log archive: [archive path], retention 7 years. Recommended preferred tier subject to UW review of inland marine sublimit, EPLI rider, and cyber liability bundle considerations. [Underwriter name, sign-date]."

The note covers all seven elements: AI acknowledgment, prompt version, model version, verification checklist, human reviewer identification, sign-date, prompt-log archive location. The note's structure flows naturally - case context first, AI involvement second, verification third, recommended decision fourth, signature fifth. The discipline produces a per-case audit-credible record without significant operational overhead.

Adjuster File-Note Pattern

The adjuster operates in Guidewire ClaimCenter or carrier-specific claim platform; file notes document the claim's progression. Sample file-note language for an AI-assisted FNOL summary and coverage analysis on the worked-example commercial property loss from Ch9-L2:

"FNOL handling AI-assisted on claim [number], commercial property loss at $850K building, 1968 construction, possible code-compliance investigation pending. Five Sigma platform generated coverage summary citing CP 00 10 10 12 Special Form, CP 11 32 10 13 Spoilage, CP 14 12 06 07 Equipment Breakdown. Verified per FNOL checklist v2.3 by [adjuster name] [date]; checklist Step 7 (coverage flags reflect actual policy provisions) identified Five Sigma output missing CP 04 05 06 13 Ordinance or Law sublimit $250K from declarations page. Adjuster override: AI output modified to add CP 04 05 06 13 with sublimit reference; reserve recommendation adjusted to reflect potential code-compliance costs within sublimit. Systematic concern escalated to claim manager [name] [date] for Five Sigma prompt-refinement consideration; subsequent vendor engagement initiated by CCO. Five Sigma prompt: claim_coverage_v4.2 (claude-3.5-sonnet-2026-05). Verification archive: [archive path], retention 7 years. HIPAA-environment: Azure OpenAI BAA confirmed (no PHI in this commercial-property scope). Initial reserves: building repair $750K + BPP $85K + code-compliance contingency within sublimit. Cause-of-loss investigation prioritized; updates pending. [Adjuster name, sign-date]."

The note documents: AI assistance (Five Sigma); platform-prompt version (claim_coverage_v4.2); model version; verification checklist execution; specific defect catch (Ordinance or Law omission); adjuster override action with rationale; systematic-concern escalation; cross-functional engagement (claim manager → CCO → vendor); HIPAA-environment confirmation. The discipline preserves the audit chain across both single-claim handling and systematic-issue escalation.

Producer Activity Log Pattern

The producer (retail or wholesale broker or captive agent) operates in Applied Epic, AMS360, or Vertafore; activity logs document client communications and submission work. Sample activity-log language for an AI-assisted broker quote comparison:

"Broker quote comparison AI-assisted via prompt producer_v4.1 (Claude 3.5 Sonnet 2026-05) for client [Synthetic Mfg LLC] renewal effective [date]. Four carrier quotes received: Hartford ($X premium with [structure]), Travelers ($Y with [structure]), CNA ($Z with [structure]), Chubb ($W with [structure]). AI generated structured comparison matrix with side-by-side coverage limits, deductibles, endorsements, premium, retention, and coverage-gap identification. Verified per broker-comparison checklist v1.4 by [producer name] [date]; principal review [name] [date]. All four carrier ratings verified current (Hartford A+ XV per A.M. Best [date]; Travelers A++ XV; CNA A XV; Chubb A++ XV). Coverage gaps identified: CNA quote excludes [specific endorsement]; Travelers quote has $50K lower cyber sublimit than Hartford. Client-facing executive summary prepared with simplified language and technical footnotes. Recommendation rationale documented. Prompt-log archive: [archive path], retention per agency records-management policy. [Producer name, sign-date]."

The activity log integrates AI documentation with producer-side discipline: rating verification per checklist; principal review where required; client-facing simplification preserved alongside technical accuracy; documented rationale for recommendation. The discipline supports producer E&O posture and carrier-relationship quality.

Actuarial File-Note Pattern

The actuary operates in reserving software (custom or vendor platforms), rate-filing systems (SERFF), and R/SAS/Python pipelines. File notes integrate with actuarial documentation under ASOP 41 communication discipline. Sample file-note language for an AI-assisted reserve memo:

"Reserve review memo for Commercial GL line, AY 2017-2025 cohort, development through 12/31/2025, AI-assisted via prompt actuary_v5.0 (Claude 3.5 Sonnet 2026-05). Source data: PolicyCenter export [date] with hash digest [hash]; ASOP 23 data-quality checks passed; reconciles to Schedule P Part 3 prior period [date]. AI generated structured triangle commentary with development-factor analysis (12-to-24 month avg 1.65, consistent with NCCI/ISO benchmarks 1.45-1.85), chain-ladder estimates on stable AYs 2017-2022, BF with prior-year ELR on AYs 2023-2024, ELR on AY 2025. IBNR estimate $52M aggregate with documented confidence intervals (±15% chain-ladder, ±25% BF, ±35% ELR). Verified per actuarial reserve-memo checklist v3.1 by [pricing actuary name] [date]; chief actuary concurrence [name] [date]. ASOP 43 unpaid-claim-estimate methodology selection per AY documented. Schedule P Part 1 reconciliation: within $0.4M variance, documented. Development-pattern anomaly investigation: AY 2022 claims emerging slightly above pattern; chief-claims-officer consultation [date] confirmed handling-practice shift in 2024; reserve strengthening $4M documented. ASOP 36 SAO supporting narrative referenced separately. ASOP 41 communications compliance attested. Prompt-log archive: [archive path], retention 7 years. [Pricing actuary name, FCAS, MAAA, sign-date]."

The note integrates AI documentation with ASOP-compliance posture: data-quality assertion (ASOP 23); methodology selection (ASOP 43); communication compliance (ASOP 41); SAO supporting-narrative reference (ASOP 36); chief actuary concurrence; chief-claims-officer consultation on development-pattern anomalies. The discipline supports the appointed actuary's signing posture on the eventual SAO and the carrier's regulatory and rating-agency dialogue.

Where to Store the Prompt Log

The prompt-log archive is the persistent storage of AI-generation records: user prompt (case-specific input), system prompt version used, model version, raw AI output, verified output post-correction, verification-checklist completion record, reviewer identification, sign-date. The 2026 mature carrier maintains the archive in one of three architectures:

Integrated PAS/claim/AMS archive. The prompt log is stored as a linked record in the policy administration system, claim system, or AMS where the artifact lives. Advantage: chain-of-custody is intrinsic; archive lives where the work lives; discovery production is straightforward. Limitation: PAS/claim/AMS storage costs may be higher than alternative architectures; some platforms have limited support for large prompt-log payloads.

Separate AI-governance archive with linkages. The prompt log is stored in a dedicated AI-governance archive (cloud storage, enterprise document management); the PAS/claim/AMS file note includes a reference link. Advantage: cost-efficient storage at scale; centralized AI-governance archive supports cross-functional and CCO QA review. Limitation: chain-of-custody requires the linkage discipline; broken links accumulate exposure.

Cloud-vendor archive with carrier access. The AI provider (Azure OpenAI, AWS Bedrock, GCP Vertex) maintains the archive within the carrier's enterprise tenancy; the carrier accesses via API. Advantage: provider-managed retention and compliance; integration with Microsoft/Amazon/Google governance frameworks; HIPAA-eligible environment by definition for PHI workflows. Limitation: vendor lock-in; cross-vendor portability considerations.

Retention is typically 7 years per carrier records-management policy for general business records; longer (10-15 years) for litigation-defense files or as state-specific retention requirements demand; permanent for catastrophic-tier artifacts (SAO, major rate filings, major treaty cessions). The retention policy is filed-equivalent infrastructure under NAIC §4 and Colorado Reg 10-1-1.

Handling Discovery Requests Including AI-Generated Content

Bad-faith litigation under the applicable state's UCSPA, ERISA §502(a), or class-action consumer-protection claims increasingly includes discovery requests for AI-generated content and supporting documentation. The 2026 plaintiff-counsel discovery patterns:

(1) Production of all AI-prompt logs related to the specific claim or coverage decision. (2) Production of carrier system prompts in effect during the relevant period. (3) Production of carrier verification checklists and their completion records. (4) Production of carrier algorithm-inventory entries and bias-test exhibits. (5) Identification of AI providers and BAA agreements. (6) Production of any cross-functional sign-off records.

The carrier's discovery cooperation framework: (1) Engage outside counsel and in-house legal early on AI-content discovery requests; (2) Preserve all responsive records per litigation-hold protocol; (3) Coordinate with CCO governance on production scope and privilege assertions; (4) Produce all responsive non-privileged records under appropriate confidentiality designations (HIPAA-protected medical content under protective order); (5) Maintain chain-of-custody on AI-prompt-log production to preserve authenticity; (6) Identify any work-product or attorney-client privileged materials for protective treatment.

The discipline parallels other discovery cooperation patterns; the AI-specific element is that AI-generated content is typically discoverable as ordinary work product, not privileged. The defense theory rests on documented governance: clean prompts, deployed checklists, complete file notes, version-stamped archives demonstrate disciplined operation; AI-content production becomes corroborating evidence of governance posture rather than incriminating evidence of negligence.

Work Product and Privilege on Defense-Panel Claims

When defense counsel is involved in a claim (typically third-party liability claims with defense-panel attorney engaged), the attorney-client privilege and work-product doctrine apply to communications between the carrier, the insured, and the defense counsel. AI involvement in a defense-panel claim introduces complexity: AI-assisted reserve analysis, AI-assisted coverage analysis, AI-assisted investigation summaries - are they privileged work product?

The 2026 emerging framework: (1) AI-assisted analysis prepared in anticipation of litigation, at the direction of counsel, and protected from disclosure to opposing parties is work product. (2) AI-assisted analysis prepared in the ordinary course of business - claim handling, coverage determination, reserve setting - is typically not work product even when AI-assisted. (3) The distinction depends on the timing and purpose of the analysis, not whether AI was involved. (4) Carrier internal communications involving AI-assisted content with defense counsel may be privileged if the communications are for the purpose of legal advice.

The carrier's discipline: when defense counsel is engaged on a claim, AI-assisted analysis intended for defense purposes is segregated with counsel-direction documentation; ordinary-course AI-assisted analysis remains in the standard work-product flow; the file note distinguishes the two categories. Mature carriers operate the segregation discipline; immature carriers may inadvertently include ordinary-course AI analysis in work-product claims and accumulate privilege-waiver risk.

File-Note Tiers and the Discovery-Cooperation Framework

File-note discipline applies across the four decision-frame tiers from Ch9-L2:

Tier 1 file note: Brief AI-assistance acknowledgment with prompt version, model version, verification checklist reference, reviewer sign-date. Single paragraph; routine documentation.

Tier 2 file note: Expanded acknowledgment including domain-reviewer sign-off; documented override actions if applicable; systematic-concern escalation if patterns surface; cross-functional engagement if needed.

Tier 3 file note: Comprehensive acknowledgment including multi-functional sign-off; CCO countersignature; legal coordination notation if applicable; chain-of-custody attestation for litigation-defense readiness.

Tier 4 file note: Strategic-level documentation including chief-officer engagement; board awareness notation if applicable; appointed actuary signature on SAO; regulatory-submission attestation; full chain-of-custody to support examination and rating-agency reviews.

The discovery-cooperation framework for a bad-faith suit on a Tier 3 artifact: (1) Litigation hold issued; (2) Tier 3 file notes and all linked prompt logs preserved; (3) Discovery production prepared in coordination with outside counsel and CCO; (4) AI-prompt logs produced under confidentiality designation; (5) Verification-checklist completion records produced; (6) Cross-functional sign-off records produced; (7) Algorithm-inventory entries referenced; (8) BAA agreements identified; (9) Deposition preparation on AI-governance discipline; (10) Trial-readiness on documented governance.

The cooperation framework operationalizes the disciplined file-note discipline at the eventual litigation moment. Carriers operating disciplined file-note discipline find their AI-content discovery production becomes corroborating evidence; carriers operating undisciplined file-note discipline find their production becomes the litigation defendant's exposure.

Key Takeaways

• File note must document AI involvement across seven elements: AI assistance acknowledgment, prompt version, model version, verification reference, human reviewer identification with sign-date, cross-functional sign-off references where applicable, prompt-log archive location. Disciplined documentation serves NAIC §4 attestation, chain-of-custody for examination/litigation, and reproducibility.
• Underwriter file-note pattern documents source documents, AI-generated structured summary content, verification checklist execution, recommended decision with rationale, underwriter signature. Operates in PolicyCenter / Duck Creek / Sapiens / Majesco with structured note fields.
• Adjuster file-note pattern documents AI platform (e.g., Five Sigma), prompt and model version, verification checklist with defect catches and override actions, systematic-concern escalation if patterns surface, HIPAA-environment confirmation for L&H/WC/BI scope, signature.
• Producer activity-log pattern in Applied Epic / AMS360 / Vertafore documents AI assistance, carrier rating verification, coverage-gap identification, client-facing simplification, principal review, recommendation rationale, signature. Supports producer E&O posture and carrier-relationship quality.
• Actuarial file-note pattern integrates ASOP discipline: ASOP 23 data quality; ASOP 43 methodology; ASOP 41 communication; ASOP 36 SAO supporting-narrative reference. Chief actuary concurrence; chief-claims-officer consultation on development-pattern anomalies; appointed-actuary sign-date.
• Prompt-log archive architecture options: integrated PAS/claim/AMS archive (intrinsic chain-of-custody); separate AI-governance archive with linkages (cost-efficient at scale); cloud-vendor archive (provider-managed compliance). Retention typically 7 years; longer for litigation-defense files; permanent for Tier 4 catastrophic artifacts.
• Discovery requests increasingly include AI-prompt logs, system prompts, verification checklists, algorithm-inventory entries, bias-test exhibits, BAA agreements, cross-functional sign-off records. Carrier cooperation framework: engage outside counsel early, preserve via litigation hold, coordinate with CCO on scope and privilege, produce under appropriate confidentiality designations.
• Work product and privilege on defense-panel claims: AI-assisted analysis at counsel direction for litigation purpose is work product; ordinary-course AI-assisted claim handling is typically not. Mature carriers segregate the two; immature accumulate privilege-waiver risk by mixing.
• File-note tiers map to four-tier decision frame: Tier 1 brief acknowledgment, Tier 2 expanded with override and escalation notation, Tier 3 comprehensive with multi-functional sign-off and CCO, Tier 4 strategic with chief-officer engagement and regulatory-submission attestation.