AI-use disclosure in SERFF filings: Washington's questions

What Washington asks

The OIC's notice adds new SERFF State Specific questions meant to capture whether AI tools were used to create or review a filing, or to develop the work it relies on. The notice describes AI tools broadly: generative AI, machine learning, or AI inside vendor tools. The work in scope includes rates, factors and rules, underwriting, policy forms, plan and product design, and networks.

There are two questions:

• Was AI used? A yes or no answer about the filing or the work it relies on.
• If yes, how? A brief description of the tool or vendor, what it did, what it affected (rates, factors, rules, forms, underwriting or network), its role and its impact. If there was no AI use, the filer enters N/A.

The notice gives illustrative answers within the character limit: one describes a general AI assistant used to draft and clarify form language, as a writing aid only, with human review; another describes a vendor tool used to check rating factors for anomalies, as a QA step before submission, with actuarial sign-off. It also says the question text is subject to change.

How much detail, and where it goes

The answer field allows 250 characters. When the answer does not fit, the notice says to attach a letter or a separate PDF on the Supporting Documentation tab. The OIC's April 2026 health and disability form filing instructions add a naming convention for that PDF, built from the company, the state, the type of insurance and the date the information was prepared; check the instructions for your line before you file. Filers may also update their answers in SERFF after submission when a correction or clarification is needed, consistent with the filing record.

On materiality, the notice asks filers to describe first the AI uses that could materially affect the substance of the filing: product or network design decisions, models, variables, assumptions and estimates, pricing factors, underwriting rules, or coverage and benefit provisions. Use only for administrative drafting or proofreading can be stated briefly.

The OIC says it will use the answers in its review of the filing and anticipates that they may reduce reviewer inquiries and review time. The questions do not change the existing review standards.

What a good answer rests on

A short, accurate answer is easier when the team knows, filing by filing, which tools touched which parts of the work. In practice that means keeping a record as you go rather than reconstructing it at submission:

• which AI tools or vendor products were used, and by whom;
• which parts of the filing they touched: rates and factors, rules, forms, underwriting material, or only drafting;
• what a person reviewed, changed or signed off afterwards;
• which files the tools produced, and where they are.

What Evermont records

When your assistant works on a filing through Evermont, every change it makes is recorded on the filing: changes to the filing's details, documents, objections and requirement rows, saved intake answers and requirement checklist builds. Each change is attributed to the signed-in person (with the assistant they used, when its name is known) or to the API key that made it.

• The Disclosure page. Each filing in the Evermont web app has a Disclosure page: the AI-use report built from that record, counting the people and API keys involved, with a JSON export.
• A draft summary. Your assistant can ask Evermont for the filing's history with a disclosure summary: a short text describing the recorded agent actions and listing the files agents saved to the filing. The filer reviews it and adapts it to Washington's questions, either as a 250-character answer or as an attached PDF.

Two limits matter. The record covers changes, not what the assistant searched or read; those appear as citations in the assistant's answers, not in the report. And Evermont sees only what happens through Evermont: AI used elsewhere, such as in a rating vendor's tool or a chat without Evermont connected, is for the filer to add. The summary Evermont drafts says so when it finds nothing.

For how the rest of the filing work runs with an assistant, see AI for insurance rate and form filings.

Questions and answers

Which filings does Washington's AI question apply to?

According to the OIC's notice, all new Washington SERFF rate and form filings from May 1, 2026, including form, rate and network filings for property and casualty and for life, health and disability. The notice says the question text may change, so check the State Specific tab on your filing.

What if AI was only used to proofread?

The notice asks filers to put the weight on uses that could materially affect the filing, such as models, variables, assumptions, pricing factors, underwriting rules or coverage provisions. If AI was used only for administrative drafting or proofreading, it says a brief statement to that effect is enough.

What if the answer does not fit in the field?

The SERFF field allows 250 characters. The notice says to attach a letter or PDF on the Supporting Documentation tab when more detail is needed, and the OIC's health and disability form filing instructions give a file naming convention for it.

Does Evermont's report cover everything Washington asks about?

No. It covers the changes agents made to the filing through Evermont, with who made them. It does not list what was searched or read, and it cannot see AI used outside Evermont, such as vendor tools or a chat without Evermont connected. The filer adds those.

Do other states ask the same question?

This page covers Washington's notice. Each state sets its own SERFF State Specific questions and instructions, so check each state's current SERFF General Instructions and filing rules before you file.