AI Usage and Human Review

Nairo uses AI to support insurance review work.

This page explains how AI is used in Nairo, where human review is required and what teams should consider before using AI-generated outputs in business workflows.

AI outputs are assistance for human review. They are not final underwriting, claims, legal, compliance or business decisions.

AI in Nairo

AI features appear across the main product surfaces:

AI is used to help users review materials faster, apply configured instructions more consistently and surface relevant information for review.

Nairo is not used to make binding insurance decisions without human oversight.

Human review requirement

Nairo is decision support software.

AI-generated outputs should be reviewed by a qualified human before they are used in business-critical workflows.

This is especially important for:

• Underwriting decisions
• Claims decisions
• Coverage-sensitive conclusions
• Policy interpretation
• Broker or client communications
• Audit preparation
• Compliance-sensitive outputs
• External distribution of AI-assisted deliverables

The appropriate reviewer depends on the customer’s internal authority model, workflow, jurisdiction and governance process.

Suggested review model

This table is guidance only. Customer teams should define their own review and sign-off policy.

Human accountability

The user remains responsible for deciding whether an AI output is appropriate to use.

Before relying on an AI output, reviewers should consider:

• Whether the correct source materials were selected
• Whether the output cites or refers to relevant source material where supported
• Whether important exclusions, limitations or exceptions were missed
• Whether the output is consistent with internal guidelines and authority limits
• Whether additional human review, legal review, actuarial review or referral is required
• Whether the output is suitable for external distribution

Nairo does not replace professional judgment, authority limits, referral processes or formal approval workflows.

Corrections, overrides and rationale

When a human reviewer disagrees with an AI output, the final decision and rationale should be recorded in the team’s governance process.

Nairo does not enforce a dedicated override registry across all surfaces today.

Some extracted values may be edited or retried where supported, and some workflows may retain basic edit or processing metadata. This should not be interpreted as a universal accept/reject/override workflow across every product surface.

For decision-memory concepts, see Decision Memory.

References and citations

Where supported, Nairo outputs may include references, citations or source passages.

References help reviewers check where an answer came from. They do not replace reading the underlying source material.

Reviewers should verify that cited sources actually support the conclusion, especially where the output affects:

• Coverage
• Appetite
• Pricing
• Claims handling
• Regulatory or compliance matters
• External communications

Citations do not guarantee completeness or correctness.

Model execution and external tools

Model execution may differ by surface and workspace configuration.

In the current implementation:

• Assistant uses backend-managed model execution
• Projects use backend-managed model execution for the Project Assistant
• Insight Table execution is backend-managed
• Experts and Actions may use client-side Gemini calls where configured

Internet search may be available for Assistant conversations where enabled. It should not be assumed to be enabled globally across all surfaces.

For more detail, see Model Providers.

What AI does not do

Nairo AI does not:

• Make autonomous accept, decline, bind, pay or deny decisions
• Guarantee that outputs are correct
• Guarantee that every relevant source has been considered
• Replace professional judgment
• Replace legal advice
• Replace actuarial sign-off
• Replace compliance review
• Replace underwriting or claims authority
• Automatically create regulatory-grade audit records
• Automatically capture override rationale across all surfaces
• Automatically turn every reviewed output into reusable institutional memory

Model governance

Workspace administrators cannot currently configure full workspace-wide model availability through a customer-facing admin UI.

Model choice, model availability and execution path may vary by surface and deployment configuration.

Teams with specific requirements around model providers, data classification, internet search, external tools or sensitive materials should confirm those requirements with Nairo before rollout.

For more detail, see Model Providers and Subprocessors.

Recommended customer policy

Before using Nairo in production workflows, teams should define:

• Which AI outputs can be used internally
• Which AI outputs require human sign-off
• Who can approve underwriting, claims, coverage or compliance-sensitive outputs
• Which materials can be uploaded or selected as context
• Whether internet search is permitted
• Which surfaces can be used with sensitive materials
• How corrections, overrides and rationale should be recorded
• Which outputs must be retained outside Nairo
• When legal, actuarial, compliance or senior review is required

Related areas

• Core Concepts
• Model Providers
• Decision Memory
• Data Handling
• Subprocessors
• Security Overview

Getting started

Define your team’s AI review policy before using Nairo for business-critical workflows.

At minimum, decide which outputs require sign-off, who can approve them and how rationale should be recorded when a human reviewer disagrees with an AI-generated suggestion.