AI in insurance, checked dailyThursday 17 September 2026
The Renewal Report
News, findings and tests. Every item with its source, its evidence and what it means for a book of business.For agencies, MGAs and carriers
Finding 001Open, revised as sources are added

What belongs in an agency AI policy, and what is actually required?

Question
Whether a retail agency is obliged to maintain a written AI policy
Short answer
No single answer. Texas expressly reaches agents; California reaches licensees; four of the six address insurers only. The detailed checklist in trade press is a recommendation, not a rule.
Basis
Five state bulletins, one adopted regulation, one policy form, two sets of professional guidance
Tier
Mixed — one regulator expectation reaching agents, the rest recommendation
Scope reviewed
TX, NY, CA, CO, WI, IL — not a fifty-state survey
Reviewed
16 September 2026

The finding

Across six states, most AI bulletins are addressed to insurers, not to producers. The obligation on a retail agency is narrower than nearly everyone assumes, and the recommendation set is considerably broader.

Texas is the exception. Bulletin B-0003-26, issued 12 June 2026, is addressed to regulated entities and their agents and representatives, and where AI makes a consequential decision it expects a person to review and agree before action is taken. It does not define "consequential decision," and it says plainly that its goal is not to prescribe specific practices or documentation requirements.

California's Bulletin 2022-5 reaches "licensees," the broadest addressee line we found in any state instrument. But it restates duties that already exist under the Insurance Code rather than creating AI-specific ones.

New York, Wisconsin and Colorado are insurer-only. Illinois addresses "regulated entities," which is broader than insurers and vaguer than helpful; whether it reaches a particular agency is a question for counsel, not for us.

So the detailed agency checklist circulating in trade press comes from defense counsel, published in a Big "I" magazine on 9 September 2026. Utica National's June 2024 E&O newsletter says agency management "must" develop a policy, that is carrier loss-prevention guidance, not a policy condition and not a statute. Both are worth following. Neither is law.

State by state

Instruments read in full for this finding
InstrumentAddressed toReaches producersTier
Texas Bulletin B-0003-26All regulated entities and their agents and representativesYes, expresslyRegulator expectation
California Bulletin 2022-5Admitted and non-admitted insurers, licensees, other interested partiesYes, via "licensees"Restates existing legal duties
Illinois Company Bulletin 2024-08All insurers and regulated entities licensed in IllinoisUnclearRegulator expectation
NYDFS Circular Letter No. 7 (2024)Insurers, Article 43 corporations, HMOs, fraternals, NYSIFNoRegulator expectation
Wisconsin OCI AI BulletinAll insurers authorized in the stateNoRegulator expectation
Colorado Reg. 10-1-1, amendedLife, private passenger auto and health benefit plan insurersNoLegal duty, on insurers

Health utilization-review restrictions in Texas (Ins. Code ch. 4201) and California (Ins. Code §10123.135) are legal duties, but they apply to utilization review and say nothing about commercial underwriting or agency work. We did not survey the other forty-four states.

The part nobody mentions

Your E&O carrier tells you to write an AI policy to manage exposure. There is at least one form on the market where having a deficient one is itself excluded.

Berkley form PC 51380 00 (06-24), Artificial Intelligence Exclusion (Absolute), clause (1)(c), excludes claims arising from inadequate or deficient policies, practices, procedures or training relating to AI, or failure to develop or implement them. The endorsement contains no safe harbor for merely maintaining a policy.

Specimen form, effective date blank, attaching to all Liability Coverage Parts

Two things this is not. It is a management liability form, so it does not establish that your professional liability policy excludes AI. And a specimen is not proof of what is attached to your program.

It is a reason to read yours. We set the Berkley definition against the new ISO general liability exclusions in Finding 002; the two are not the same width.

What to do with this

Write the policy anyway. The recommendation set, approved tools, permitted data, licensed review before advice, authority boundaries, source verification, testing, error escalation, documentation, is sound risk management whether or not a regulator requires it, and in Texas the review expectation reaches you directly.

Then read the exclusions on every liability form in your own program, not the specimen. An eight-clause outline with the basis and tier of each clause is in the library, labeled as ours.

Sources: Texas TDI Bulletin B-0003-26 (issued 12 June 2026; web page last updated 16 July 2026); California CDI Bulletin 2022-5 (30 June 2022); Illinois DOI Company Bulletin 2024-08 (13 March 2024); NYDFS Insurance Circular Letter No. 7 (11 July 2024); Wisconsin OCI bulletin (18 March 2025); Colorado Regulation 10-1-1 as amended (adopted 20 August 2025); Ventura & Gittleman, Independent Agent magazine (9 September 2026); Utica National E&O Risk Management Newsletter, Vol. II Issue 6 (June 2024); Berkley PC 51380 00 (06-24) specimen. The NAIC model bulletin itself could not be retrieved; NAIC-derived language was read through the Wisconsin and Illinois instruments.