The short answer
If you run marketing, growth or digital at an insurer, bank, credit union or asset manager in the Hartford area, the AI questions you get asked are not the ones in the vendor decks. They come from the general counsel, the chief risk officer and the board, and they are about governance: which of our AI tools does the regulator care about, who signs what, what do we owe them about our vendors, and what happens to all of this in a hiring or restructuring decision. This piece answers those questions in the order they usually come up, using the two documents that actually apply in Connecticut.
The first is Connecticut Insurance Department Bulletin MC-25, issued February 26, 2024, which sets out what the Department expects from every insurer licensed in the state that uses AI, and names marketing as one of the functions in scope. The second is Public Act 26-15, the state’s 2026 AI law, whose employment provisions apply to every Connecticut employer, regulated or not. We summarised that law for mid-market companies in Connecticut’s new AI law: what a mid-market company has to do, and by when; this piece picks up the parts that land on a financial-services company.
The usual caution applies. We are not lawyers, and this is not legal advice. It is a plain-language reading of published documents, written for the executives who will have to decide what to change. Check anything you act on against the original text and with your counsel and compliance team.
Does the insurance bulletin cover our marketing team, or just underwriting and claims?
It covers marketing. The bulletin’s scope sentence lists the insurance practices where AI use is expected to comply with the law, “including product development, marketing, sales and distribution, underwriting and pricing,” and goes on to claims and fraud detection. The guideline describing what a written AI Systems Program should cover repeats the list. So a model that decides which prospects see which product, a tool that scores leads for an agent to call, or a system that personalises an offer is inside the programme, not beside it.
What the bulletin is actually worried about is an “Adverse Consumer Outcome”: a decision affecting a consumer that is subject to the Department’s rules on unfair trade practices and unfair discrimination. The test to apply to any marketing tool is whether its output changes which consumers get what, at what terms. A drafting assistant that helps a copywriter does not make a regulated decision. A model that quietly suppresses an offer to one segment does, whatever the vendor calls it. Both belong in the inventory; only one needs the full control set.
The regulator does not ask whether marketing uses AI. It asks whether you can show who the model affected, how you checked it, and who was accountable.
Who has to sign what, and when?
Every licensed insurer: a written AIS Program
The bulletin asks insurers to “develop, implement, and maintain a written program” for the responsible use of AI systems that make or support regulated decisions, scaled to the risk to consumers. It expects senior management to be “accountable to the board or an appropriate committee of the board” for the programme and for the company’s AI strategy. That is a named executive and a standing agenda item, not a policy PDF.
Connecticut-domiciled insurers: an annual certification
Insurers organised in Connecticut file an Artificial Intelligence Certification, due September 1 each year, signed by an authorised officer. It asks the company to certify either that it uses no covered AI systems, or that its use is substantially consistent with the bulletin because it has an AIS Program, is establishing one, or can show compliance by other means, with an explanation attached. Connecticut kept this certification even though the national model bulletin it follows does not require one.
Everyone: be ready to produce the records
Whether or not a written programme exists, the bulletin says the Department may ask during an examination for the programme itself, governance and training materials, data governance and bias analysis, model inventories, documentation, validation and testing, evidence of monitoring for model drift, and third-party due diligence, contracts and audits. If marketing cannot produce its part of that list, the gap is marketing’s.
Consumers: notice that AI is in use
The programme guidelines include “addressing the provision of notice to affected consumers that AI Systems are in use.” For a marketing team this usually means the privacy notice, the web forms and the chat experiences, which are the surfaces the team already owns and the ones nobody else will think to update.
Our marketing vendors all use AI now. What do we owe the regulator about them?
More than most teams assume. The bulletin’s third-party section expects an insurer to assess the data and AI systems of its vendors to confirm they meet the insurer’s own legal obligations, to include, where available, contract terms that give audit rights and require the vendor to cooperate with regulators, and to confirm compliance through audits or other checks. In practice the marketing stack is where this bites hardest, because it is bought piecemeal, often on a card, by people who were never told it was in scope.
The working list is short and tedious: every tool that touches a consumer decision, what model it runs, what data it sees, who at the vendor can answer questions, and whether the contract has the clauses. Ask each vendor, in writing, what they will provide for your AIS Program. The ones who answer quickly are the ones to keep. This is the same discipline we found in the first audit at seven credit unions, banks and insurers we have worked with, where compliance has to be a design constraint at the brief rather than a review step at the end.
/ 05We are a bank, a credit union or an asset manager, not an insurer. Does any of this reach us?
The insurance bulletin does not. Your AI expectations come from your own regulators, and we are deliberately not summarising those here. Two things reach you anyway. Public Act 26-15 applies to every Connecticut employer: from October 1, 2026, a layoff notice under the state’s plant-closing law has to say whether AI or other technology is behind the cuts, and “the software made the call” stops being a defence to a discrimination complaint; from October 1, 2027, AI tools that are a substantial factor in hiring, promotion or discipline need a written notice to the person first. And the advertising rules that already govern your marketing, the ones that shape rate claims, disclosures and data handling, do not care whether a person or a model wrote the copy. A generated ad is still your ad.
/ 06What should we ask an AI consultant before hiring one in Hartford?
Ask for the inventory first. A consultant who cannot produce a plain list of the AI tools in your marketing and growth functions, what each one decides and who owns it, is selling you a pilot, not a programme. Ask how their recommendations map to the four areas of an AIS Program, because the board will ask the same thing. Ask what they will not do: an AI consultant is not your counsel and should not validate your pricing models, and any firm that implies otherwise is the wrong firm. And ask who else they have done this for in a regulated business, then check the references properly.
If the work touches underwriting, claims, pricing or any customer decision, the hiring conversation should also settle five things in writing, because the Department can ask for all of them. First, a model-risk note for anything the engagement recommends: what the model decides, what data it was built on, how it was validated and who re-checks it. Second, vendor diligence on any third-party AI system or data source the consultant proposes, since the bulletin puts responsibility for that diligence on the insurer, not the vendor. Third, contract language with those vendors that gives you audit rights and requires cooperation with a regulatory inquiry; a SOC 2 report is useful evidence of a vendor’s controls, but it is not a substitute for those terms. Fourth, a record trail: the inventory entry, the test results and the sign-off, kept where a market conduct examiner could read them. Fifth, independence: confirmation that the consultant earns nothing from the tools they recommend. None of this is exotic. It is the ordinary paperwork of a regulated business, written down before the pilot starts rather than reconstructed after a records request.
Then ask about the opportunities, because governance on its own is a cost. The reason to do the inventory properly is that it is also the map of where AI can take work off the team, shorten the time from a question to an answer, and make marketing’s numbers something finance believes. That is why our AI Opportunity Audit starts with the inventory in week one and puts governance next to the opportunities rather than after them. It runs three weeks at a fixed fee, and the AI consulting in Connecticut page explains how those engagements work here. If you are comparing options, who shows up when you search for an AI consultant in Connecticut maps the market honestly, and how much an AI consultant costs gives you the numbers to compare against.
/ 07Questions people ask about AI governance in Connecticut financial services
What is Connecticut Insurance Department Bulletin MC-25?
A bulletin issued February 26, 2024, setting out the Department’s expectations for how insurers licensed in Connecticut govern their use of AI systems, and what AI-related information the Department may ask for during a market conduct investigation or examination. It replaced the Department’s April 2022 notice on big data and follows the NAIC model bulletin adopted in December 2023.
Does the bulletin apply to marketing?
Yes. Marketing is named in the bulletin’s list of insurance practices in scope, alongside product development, sales and distribution, underwriting, pricing, claims and fraud detection.
When is the Connecticut AI certification due?
On or before September 1 each year, for insurers domiciled in Connecticut. The first was due September 1, 2024. It is signed by an authorised officer and asks the insurer to certify that it uses no covered AI systems, or that its use is substantially consistent with the bulletin.
What does an AIS Program have to contain?
The bulletin describes four areas: the general scope of the programme across the AI lifecycle and all insurance practices; governance, including documentation standards and lines of accountability; risk management and internal controls, including data quality, bias analysis, model inventories, validation, testing and record retention; and oversight of third-party AI systems and data.
Does Connecticut’s 2026 AI law apply to banks and credit unions?
Its employment provisions apply to every Connecticut employer. The AI line in layoff notices and the removal of the “AI decided” defence take effect October 1, 2026; the written notice before AI-influenced hiring, promotion or discipline decisions takes effect October 1, 2027.
Do we need SOC 2 reports and audit rights from our AI vendors?
The bulletin expects an insurer to do its own due diligence on third-party AI systems and data, and to include terms in vendor contracts, where appropriate, that give the insurer audit rights and require the vendor to cooperate with a regulatory inquiry. It does not name SOC 2. A SOC 2 report is a sensible piece of that diligence, especially for tools that handle customer data, but the contract terms and your own written review are what the Department will ask to see.
Related reading: Connecticut’s new AI law for mid-market companies, should marketing own AI strategy?, and what happens in a three-week AI readiness audit.
This piece is one stage of a longer path. Getting AI to run inside a mid-market company puts all thirty-six of the Signal’s AI-adoption questions in the order they come up, with the governance questions in their place.