Industries · Insurance

The law calls you a Licensee. The AI bulletin tells examiners what to ask for.

A certificate of authority is examined state by state. The NAIC AI bulletin and NYDFS Circular Letter No. 7 reach the code behind pricing, underwriting, and claims. Agents now change that code.

Pre-production only: how agents change your systems, not how those systems run.

  • Chief Risk Officer
  • Head of Model Risk
  • Chief Actuary
  • CISO

Session replay

Illustrative · demo data

Session replay with demo data: an agent run with its edits, the approval it waited for, and the verified result.

Every agent run on pricing or claims code replays with its edits and approvals.

Industry trend

Source: Stack Overflow Developer Survey 2026, self-reported; data under ODbL 1.0. AutoDevOps is not affiliated with Stack Overflow.

01Where agents meet your obligations

Three sessions your examiner will ask about.

  1. 01

    An agent refactors a rating factor in the auto pricing module.

    An independent approver signs off. With no recorded reason for the change, a second finding opens.

    INS.PRICING-REVIEW
  2. 02

    An agent edits the system prompt of the claims intake assistant.

    It needs an independent approver and counts toward change velocity in the underwriting packet.

    BASE.AGENT-SYSTEM-REVIEW
  3. 03

    An agent builds a test fixture from a policyholder file and calls an unapproved model.

    A critical finding names the provider and cites the data security model law.

    INS.NPI-PROVIDERS

02The authorization at stake

If the rating engine changed and no one can say why, the rate is not on file.

The NAIC AI bulletin stands on the Unfair Trade Practices Act. That Act suspends licenses.

At stake
Certificate of authority in every state you write in. The Insurance Data Security Law calls you a Licensee.

  1. 01

    Market conduct action

    Examiners ask how your AI systems change. An unexplained rating change is what they find.

  2. 02

    Special report or examination in New York

    DFS can examine AI use or demand a special report.

  3. 03

    Cease-and-desist and penalty order

    After a hearing: penalties, or suspension or revocation of the license.

  4. 04

    Data security penalties and the April certification

    Policyholder data at an uncontracted provider is a Licensee’s failure.

Sources: NAIC Model Bulletin §4 · NYDFS Circular Letter No. 7 (2024) ¶9, ¶29 · NAIC Model #880 §8.A(2), §11.B · NAIC Model #668 §4.F, §10 · Directive 2009/138/EC (Solvency II), Art. 144

Suspension is rare and follows a hearing. Penalties and rate refilings are not, and each starts with a change nobody can explain. The record is evidence for your examiner, not a finding of compliance.

03Obligations checked

Every clause, checked on every session.

Review and forbidden-action rules hold the agent before it acts. Provider, data, and intent rules record after.

  1. BASE.AGENT-SYSTEM-REVIEW

    Agent prompts, tools, and guardrails get an independent reviewer

    Internal policy · AI system change policy · Customer AI system change policy; EU AI Act Art. 14 (human oversight)

    EUAI.ART14-03ISO42001.RSK-01NIST.MANAGE-04

  2. BANK.NYDFS-APPSEC

    Agent pushes to release branches get an independent approver

    Regulation · 23 NYCRR Part 500 · 23 NYCRR 500.8 (application security)

    NYDFS.APP-01FFIEC.SOD-01

  3. INS.PRICING-REVIEW

    Rating and underwriting code gets an independent reviewer

    Regulation · NAIC Model Bulletin (Dec. 2023): AIS Program risk management and internal controls

    NAIC.CTRL-02FFIEC.SOD-01

  4. INS.PRICING-INTENT

    Pricing changes carry documented intent

    Regulation · NYDFS Insurance Circular Letter No. 7 (2024): governance and documentation

    NAIC.GOV-01EUAI.ART11-01

  5. INS.NPI-PROVIDERS

    Policyholder data reaches covered providers only

    Regulation · NAIC Insurance Data Security Model Law (#668) · Model Law #668 §4.F (oversight of third-party service providers)

    NAIC.TP-03NYDFS.TPSP-03

Plus 4 baseline obligations every pack checks
  1. BASE.SECRET-CONTEXT

    Credentials stay out of agent context

    Framework · ISO/IEC 27001:2022 · Annex A 5.17 (authentication information)

    ISO27001.DAT-02EUAI.ART15-04

  2. BASE.DPA-PROVIDERS

    Model calls stay on the DPA subprocessor list

    Contract · Customer data processing agreement · DPA subprocessor schedule

    DORA.TPRM-02NAIC.TP-03NYDFS.TPSP-03

  3. BASE.PROD-CHANGE-REVIEW

    Infrastructure changes get an independent approver

    Framework · SOC 2 Trust Services Criteria · CC8.1 (change management)

    SOC2.CC8-03FFIEC.SOD-01DORA.CHG-04

  4. BASE.NO-FORCE-PUSH

    Agents never rewrite shared history

    Internal policy · Secure development standard · Protected branch policy

    FFIEC.LOG-03MAS.LOG-04

Templates ship without dollar figures; the register shows a labeled illustrative band until you add yours. Pre-run checks use review mode, a customer-cloud CLI capability.

05For your risk and insurance reviewers

Evidence an AI liability underwriter can rate.

Verified controls
Gate coverage, independent review rate, and self-approvals.
Change velocity
Weekly changes to prompts, tools, or guardrails, and the share with a stated reason or approval.
Autonomy
Unattended sessions and the allow, hold, and block mix.
Obligation posture
Findings by status, priced per obligation.

Production behavior is out of scope and marked that way. The packet is evidence for a reviewer. It is not insurance, a certificate of coverage, or a premium quote.

Sources and citations · 6 rules referenced

Bulletin adoption and wording vary by state, and so do the penalty statutes Model #668 and #880 point to. Your compliance team confirms which rules apply and whether coding agents sit inside your AIS Program scope. For your counsel and compliance team, not legal advice.

Other industries: Banking, Healthcare, Payments.

Bring one insurance obligation

We run it against your agents’ recorded sessions and hand you the findings, priced exposure, and evidence packet.