AI agents in insurance
Where customer, broker and carrier agents could exchange quotes and claims on ACORD and CSIO data, and the US and Canadian insurance rules that apply to AI.
Insurance runs on requests between three kinds of party: the customer, the intermediary (a broker or agent), and the carrier. A quote needs risk details from the customer and rates from carriers. A claim needs a first notice, documents, an adjuster’s decision and a payment. Carriers and brokers have published data standards for this exchange. Customers usually get web forms and phone lines. An agent acting for a customer, and one acting for a broker or carrier, could exchange the same fields directly, with people kept for coverage decisions and disputed claims.
Where agents meet today
Carrier and broker data standards
ACORD publishes insurance data standards in four families:
| ACORD standard family | What it covers, per ACORD |
|---|---|
| Property and casualty | XML and AL3 standards, also used for delegated authority and workers’ compensation |
| Life and annuity | XML standards, plus DTCC EDI specification files |
| Global reinsurance and large commercial | XML and JSON standards for claims, placing, and accounting and settlement |
| Next-Generation Digital Standards | Granular, transaction-centric standards for APIs, microservices and IoT |
In Canada’s broker channel, the Centre for Study of Insurance Operations (CSIO) publishes CSIO Data Standards and CSIO eDocs for property and casualty insurers and brokers.
These standards give agents a shared vocabulary. A customer’s agent that asks for a quote with ACORD- or CSIO-shaped risk data is asking in a form the carrier’s systems already understand.
Customers
On the customer side, quotes and claims start through web forms, apps, calls to a broker, or a carrier’s contact centre. A customer’s agent using those channels has to fill forms written for people, which is where a direct, structured exchange saves the most rework.
What a structured claim exchange looks like
Illustrative: after a minor car accident, a customer’s agent reports the loss to the carrier’s agent with the policy number, date, location, a short description and photos as file parts. The carrier’s agent opens a claim, returns the claim number, and asks for the other driver’s details as named fields. Two days later the customer’s agent asks for status and gets back the adjuster’s name, the next step and any documents still missing. A decision on coverage stays with the adjuster.
Top use cases
| Use case | Who talks to whom | Notes |
|---|---|---|
| Claims and quotes | Customer agent and carrier or broker agent | First notice of loss, claim status, document requests, quote comparison |
| Broker and carrier exchange | Broker agent and carrier agent | Submissions, quotes and subrogation between businesses |
| Bill negotiation | Customer agent and carrier agent | Renewal pricing and coverage changes |
| When to keep a human | Either side and a person | Coverage disputes, claim denials, vulnerable customers |
Claim status is the easiest start: it is read-only, frequent, and every answer maps to a claim number both sides already have.
Regulatory considerations
Insurance is regulated by states in the US and mostly by provinces in Canada, with federal prudential oversight of some insurers. Checked on 26 September 2026. This is not legal advice.
United States
- State regulation, coordinated by the NAIC. Each state’s insurance department regulates carriers and producers. The National Association of Insurance Commissioners writes model laws and bulletins that states can adopt.
- NAIC AI Model Bulletin (December 2023). The bulletin sets the expectation that decisions or actions made or supported by AI comply with all applicable insurance laws, and it sets governance expectations for how insurers oversee their AI systems. An agent that quotes or triages claims for an insurer would fall inside that governance.
- More NAIC work is in progress. According to the NAIC’s AI page (updated 3 April 2026), an AI Systems Evaluation Tool was being piloted by states as of March 2026 with adoption expected at the Fall 2026 National Meeting, and a working group is developing a framework for third-party data and models used by insurers. Neither was final when checked.
- Claims handling. The NAIC’s Unfair Claims Settlement Practices Act, a model for state law, treats failing to acknowledge claim communications with reasonable promptness, and refusing to pay claims without a reasonable investigation, as unfair claims practices. If a carrier accepts claim messages from agents, those messages are claim communications.
Canada
- Provincial conduct regulators. Market conduct is provincial: for example FSRA in Ontario and the Autorité des marchés financiers (AMF) in Québec. FSRA uses the Guidance on Conduct of Insurance Business and Fair Treatment of Customers, adopted jointly by the Canadian Council of Insurance Regulators and the Canadian Insurance Services Regulatory Organizations on 27 September 2018, to supervise insurer conduct.
- AMF AI guideline. The AMF’s Guideline for the Use of Artificial Intelligence applies to any use of AI systems by the institutions it covers, including authorized insurers, and takes effect on 1 May 2027.
- OSFI model risk (federal). OSFI’s Guideline E-23 on model risk management, published 11 September 2025 and effective 1 May 2027, applies to federally regulated financial institutions including life and property and casualty insurers, and explicitly addresses AI and machine learning models.
- Automated decisions in Québec. Section 12.1 of Québec’s private-sector privacy act requires a business that makes a decision based exclusively on automated processing of personal information to tell the person, explain the main factors on request, and let the person submit observations to someone who can review the decision.
- Federal privacy reform. Bill C-36, introduced on 15 June 2026 to replace Part 1 of PIPEDA, is at second reading in the House of Commons.
Where to start
- Start with claim status. Read-only, keyed to a claim number, and useful to both sides.
- Speak the existing data language. Map agent inputs and outputs to ACORD or CSIO fields so carriers and brokers can process them without translation.
- Treat the agent as an AI system under governance. Inventory it, classify its risk and document its limits, as the NAIC bulletin, the AMF guideline and OSFI E-23 expect for AI used by insurers.
- Verify who is asking. A claim status answer discloses personal information. Check which agent is calling and whom it represents before answering, the practice the industry calls Know Your Agent.
- Keep decisions with people where required. Coverage denials and disputed valuations should reach an adjuster with the full exchange attached, and in Québec, section 12.1 requires a route to a person who can review a fully automated decision.
Sources
- ACORD Data Standards (accessed )
- Centre for Study of Insurance Operations (CSIO) (accessed )
- NAIC: Artificial Intelligence (insurance topic page) (accessed )
- NAIC Model 900: Unfair Claims Settlement Practices Act (accessed )
- FSRA: Fair Treatment of Customers in Insurance (accessed )
- AMF: Guideline for the Use of Artificial Intelligence (accessed )
- OSFI: Guideline E-23, Model Risk Management (2027) (accessed )
- LégisQuébec: Act respecting the protection of personal information in the private sector (section 12.1) (accessed )
- LEGISinfo: Bill C-36 (45-1), Protecting Privacy and Consumer Data Act (accessed )