ClovieClaims and underwriting agents, with a person in the loop and a record of every decision.
Hold claim payouts for an adjuster, keep protected attributes out of decisions, give every agent a risk level and show regulators which AI systems touched which decision.
Claims agents
| Policy | Decision | Mode |
|---|---|---|
| Approve claim payouts | Approval | Enforce |
| No protected attributes | Deny | Enforce |
| A risk level for every agent | Deny | Enforce |
| Structured output for decisions | Deny | Enforce |
| Oversight for high-risk agents | Approval | Monitor |
| Personal data out of prompts | Deny | Enforce |
- Tool
- claims.issue_payout
- Claim
- Water damage
- Requested
- 4 min ago
Every AI-assisted decision has to stand up to scrutiny.
Agents that pay out
Claims agents can issue payouts and refunds through the same tools your adjusters use.
Unfair outcomes
Protected attributes, and data that stands in for them, can drift into pricing and claims decisions.
Proof on request
Regulators ask which AI systems an insurer uses, how they are tested and who answers for them.
The rules insurers are working to
The rules and guidance that shape how AI is controlled in insurance, each linked to its official source.
- NAIC model bulletin on AINational Association of Insurance CommissionersExpects a written AI program with governance, risk management, testing and third-party oversight. (official source, opens in a new tab)
- NYDFS Circular Letter No. 7 (2024)New York Department of Financial ServicesUse of AI systems and external consumer data in underwriting and pricing, including fairness testing. (official source, opens in a new tab)
- Colorado SB21-169Colorado General AssemblyRestricts insurers' use of external consumer data and algorithms that unfairly discriminate. (official source, opens in a new tab)
- EU AI Act, Annex IIIEuropean Union, Regulation (EU) 2024/1689Treats risk assessment and pricing for life and health insurance as high-risk uses of AI. (official source, opens in a new tab)
- EIOPA opinion on AI governanceEuropean Insurance and Occupational Pensions AuthoritySupervisory expectations for AI governance and risk management in insurance. (official source, opens in a new tab)
- OSFI Guideline E-23Office of the Superintendent of Financial InstitutionsModel risk management for federally regulated insurers and banks, covering AI and machine learning. (official source, opens in a new tab)
Clovie maps its controls to these texts to help your teams; it is not legal advice, and using Clovie does not on its own make an organization compliant with any of them.
Controls for claims and underwriting
6 controls, each a policy you install from the Policy Hub. Every one starts in monitor mode and changes nothing until you promote it after a simulation on your own traffic.
From the Policy Hub
Six policies from the action governance, fairness, transparency, data privacy and EU AI Act packs.
- A person approves
Approve claim payouts
A person approves every payout, payment or refund an agent starts, with the run trace attached.
Agent action governance pack - Blocked
No protected attributes
Blocks prompts that use a protected attribute such as race, religion or disability as a decision criterion.
Fairness and bias pack - Blocked
A risk level for every agent
Blocks a registered agent that has no risk level, so every claims and underwriting agent is classified.
EU AI Act pack · mapped to Article 6, classification - Blocked
Structured output for decisions
Agents with financial, medical or legal decision permissions must use a model that supports tools or structured output.
Transparency pack - A person approves
Oversight for high-risk agents
A person approves requests from high and critical risk agents in production before they reach a model.
EU AI Act pack · mapped to Article 14, human oversight - Blocked
Personal data out of prompts
Blocks prompts with social security, card, passport or licence numbers and dates of birth.
Data privacy pack
How a claim payout an agent starts is governed.
Where your policyholder data lives
Run Clovie as SaaS, or install it in your own cloud account or data center so claims files, prompts and traces stay where your data residency rules say.
- SaaS
- Customer VPC
- On-premises
- Air-gapped
Your cloud or data center
Self-hosted installs run the gateway, policy engine and console in your own account.
Approved vendors only
Limit claims and underwriting agents to the model providers your vendor review approved.
Evidence for examiners
Export decision logs and evidence packs when a regulator asks how a decision was made.
Controls mapped to the frameworks you answer to
- NIST AI RMF: NIST AI Risk Management Framework
- EU AI Act: EU Artificial Intelligence Act
- ISO/IEC 42001: ISO/IEC 42001 AI management system
- OSFI E-23: OSFI Guideline E-23, model risk management (Canada)
Questions
Is there an insurance policy pack?
The controls on this page come from packs that apply across industries: action governance, fairness and bias, transparency, data privacy and the EU AI Act pack. Install the ones you need and scope them to claims, underwriting or service agents.
How does this help with the NAIC model bulletin?
The bulletin expects an AI program with governance, risk management, testing and third-party oversight. Clovie keeps the inventory and owners current, enforces approval and fairness controls at the gateway and exports the decision record. It is one part of your program, not the whole of it.
Can an adjuster see why an agent proposed a payout?
Yes. The approval carries the run trace: the prompts, the tool calls and the policies that were evaluated, so the adjuster decides with the full context.
Are Canadian insurers covered by E-23?
OSFI Guideline E-23 applies to federally regulated insurers as well as banks. Install the OSFI E-23 model risk pack for its inventory, risk rating, review and deployment controls.
Will these policies block anything on day one?
Every policy installs in monitor mode, records what it would have done, and changes nothing until you promote it.
See claims and underwriting controls on your own agents.
We'll run the controls on your own agents in monitor mode and show what they would have held and blocked.